Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

bisect prints " is the first bad commit" · petrkindlmann | DecimalAI

Testingby petrkindlmann

bisect prints " is the first bad commit"

Turn a vague bug report into a VERIFIED minimal reproduction and then a failing regression test, agent-driven end to end. Covers extracting the implicit repro from a thin report (env, build, steps, data), the reproduce-minimize-isolate-capture loop, git bisect to find the introducing commit, building a deterministic minimal repro (fixed seeds, frozen time, stubbed network), writing the failing regression test BEFORE the fix (red) and confirming the fix flips it green, and writing repro evidence

Maintained

SkillScore

SkillScore95/100

Test results 95

Safety

3/3 safety checks passed

Code scan
Intent
Content

What it changes

73%95%of test cases passed +23 pts

Without skill16 of 22

With skill21 of 22

6 fixed15 passed either way0 still failing1 broke

Measured on gemini-3.8-flash11/2/2026v1

Track record

Agent runs: 0; All time

See one task, with and without the skill

One test taskA QA tester handed over a 15-step browser recording showing an item price discrepancy that occurs intermittently during late-night flash sales with dynamic promo codes. How should we systematically pare this down into the smallest possible reproduction flow, lock down any sources of flakiness, and prove it triggers reliably across multiple automated executions?

Without the skillMissed

With the skillPassed

Same model, same task. Answers are capped and redacted.

Use this skill

Your agent runs on

Adds every file the skill has, pinned so you can update it. No account needed.

pip install "decimalai>=0.15"
decimalai skills add petrkindlmann-bug-reproduction
claude --plugin-dir "$(git rev-parse --show-toplevel 2>/dev/null || pwd)/.agents"

The same install from Python, for a notebook or a setup script.

Install

pip install "decimalai>=0.15"

Add the skill

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

print("Added. Start Claude Code with the command below.")

Start Claude Code with it

claude --plugin-dir "$(git rev-parse --show-toplevel 2>/dev/null || pwd)/.agents"

Type these in a Claude Code session. No account or API key. It writes the skill and any files it bundles, and asks first when its safety rating isn't Passed.

Add the DecimalAI plugin, once

/plugin marketplace add decimal-labs/decimalai-claude-plugins
/plugin install decimalai@decimalai

Install this skill

/decimalai:install petrkindlmann-bug-reproduction

The plugin needs python3 and curl on your machine.

No Python or account. Claude Code picks it up in its next session.

Save it into your project

curl -fsSL --create-dirs -o .claude/skills/petrkindlmann-bug-reproduction/SKILL.md https://app.decimal.ai/s/petrkindlmann-bug-reproduction/SKILL.md

Copy SKILL.md

Save it in your project, as

.claude/skills/petrkindlmann-bug-reproduction/SKILL.md

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Claude Code docs

agent_config() hands your agent the skill. With no DecimalAI key it reads your project's files and sends nothing. Needs SDK 0.27 or newer.

Install

pip install "decimalai[claude-agent-sdk]"
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

import asyncio

from claude_agent_sdk import ClaudeAgentOptions
from decimalai.claude_agent_sdk import agent_config

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

async def main():
    question = "Hello!"
    async with support.run(question) as run:  # one user message, one run
        async for message in run.query(options=ClaudeAgentOptions(max_turns=6)):
            print(message)

# In a notebook or async code: await main()
asyncio.run(main())

Prefer its own loader? ClaudeAgentOptions(plugins=[skills_plugin()], skills="all") reads the same folder. Use one or the other: a skill delivered both ways reaches the prompt twice.

The same install from Python, for a notebook or a setup script. agent_config() hands your agent the skill. With no DecimalAI key it reads your project's files and sends nothing. Needs SDK 0.27 or newer.

Install

pip install "decimalai[claude-agent-sdk]"

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

import asyncio

from claude_agent_sdk import ClaudeAgentOptions
from decimalai.claude_agent_sdk import agent_config

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

async def main():
    question = "Hello!"
    async with support.run(question) as run:  # one user message, one run
        async for message in run.query(options=ClaudeAgentOptions(max_turns=6)):
            print(message)

# In a notebook or async code: await main()
asyncio.run(main())

Prefer its own loader? ClaudeAgentOptions(plugins=[skills_plugin()], skills="all") reads the same folder. Use one or the other: a skill delivered both ways reaches the prompt twice.

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

import asyncio
from pathlib import Path
from claude_agent_sdk import ClaudeAgentOptions, query

skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
options = ClaudeAgentOptions(system_prompt=f"You are a helpful agent.\n\n{skill}", max_turns=6)

async def main():
    async for message in query(prompt="Hello!", options=options):
        print(message)

# In a notebook or async code: await main()
asyncio.run(main())

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Claude Agent SDK docs

agent_config() hands your agent the skill. With no DecimalAI key it reads your project's files and sends nothing. Needs SDK 0.27 or newer.

Install

pip install "decimalai[openai-agents]"
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

from agents import Agent, Runner
from decimalai.openai_agents import agent_config

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

agent = Agent(
    name=support.name,
    instructions=support.instructions,  # your prompt, then the skill list
    tools=[*support.tools],             # your tools, then load_skill
)
# In a notebook or async code: result = await Runner.run(agent, "Hello!")
result = Runner.run_sync(agent, "Hello!")

Using SandboxAgent? capabilities=[Shell(), decimalai.openai_agents.sandbox_skills()] points its own loader at the same folder.

Install

pip install "decimalai[openai-agents]"

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

from agents import Agent, Runner
from decimalai.openai_agents import agent_config

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

agent = Agent(
    name=support.name,
    instructions=support.instructions,  # your prompt, then the skill list
    tools=[*support.tools],             # your tools, then load_skill
)
# In a notebook or async code: result = await Runner.run(agent, "Hello!")
result = Runner.run_sync(agent, "Hello!")

Using SandboxAgent? capabilities=[Shell(), decimalai.openai_agents.sandbox_skills()] points its own loader at the same folder.

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
from agents import Agent, Runner

skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
agent = Agent(name="support", instructions=f"You are a helpful agent.\n\n{skill}")
# In a notebook or async code: result = await Runner.run(agent, "Hello!")
result = Runner.run_sync(agent, "Hello!")

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

OpenAI Agents docs

agent_config() hands your agent the skill. With no DecimalAI key it reads your project's files and sends nothing. Needs SDK 0.27 or newer.

Install

pip install "decimalai[openai]"
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

import json

from openai import OpenAI
from decimalai.openai import agent_config

client = OpenAI()
# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")


def lookup_order(order_id: str) -> str:
    return f"Order {order_id} shipped yesterday."


# Your tools, as Chat Completions takes them, and what answers each.
TOOLS = [{"type": "function", "function": {
    "name": "lookup_order",
    "description": "Look up the status of an order.",
    "parameters": {"type": "object", "properties": {"order_id": {"type": "string"}},
                   "required": ["order_id"]},
}}]
FUNCTIONS = {"lookup_order": lookup_order}

question = "Where is order A1001?"
with support.run(question) as run:  # one user message, one run
    # run.system is your prompt, then the skill list.
    messages = [{"role": "system", "content": run.system}, run.message]
    while True:
        reply = client.chat.completions.create(
            model="gpt-5.4-mini",
            messages=messages,
            tools=[*TOOLS, *run.tools],  # your tools, then load_skill
        ).choices[0].message
        messages.append(reply)
        if not reply.tool_calls:
            break
        for call in reply.tool_calls:
            # load_skill is answered by the run; your own tools by FUNCTIONS.
            function = FUNCTIONS.get(call.function.name)
            messages.append(run.tool_result(call) or {
                "role": "tool", "tool_call_id": call.id,
                "content": function(**json.loads(call.function.arguments)) if function
                else f"There is no tool named {call.function.name!r}.",
            })
print(reply.content)

# The Responses API takes run.response_tools instead of run.tools.

Install

pip install "decimalai[openai]"

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

import json

from openai import OpenAI
from decimalai.openai import agent_config

client = OpenAI()
# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")


def lookup_order(order_id: str) -> str:
    return f"Order {order_id} shipped yesterday."


# Your tools, as Chat Completions takes them, and what answers each.
TOOLS = [{"type": "function", "function": {
    "name": "lookup_order",
    "description": "Look up the status of an order.",
    "parameters": {"type": "object", "properties": {"order_id": {"type": "string"}},
                   "required": ["order_id"]},
}}]
FUNCTIONS = {"lookup_order": lookup_order}

question = "Where is order A1001?"
with support.run(question) as run:  # one user message, one run
    # run.system is your prompt, then the skill list.
    messages = [{"role": "system", "content": run.system}, run.message]
    while True:
        reply = client.chat.completions.create(
            model="gpt-5.4-mini",
            messages=messages,
            tools=[*TOOLS, *run.tools],  # your tools, then load_skill
        ).choices[0].message
        messages.append(reply)
        if not reply.tool_calls:
            break
        for call in reply.tool_calls:
            # load_skill is answered by the run; your own tools by FUNCTIONS.
            function = FUNCTIONS.get(call.function.name)
            messages.append(run.tool_result(call) or {
                "role": "tool", "tool_call_id": call.id,
                "content": function(**json.loads(call.function.arguments)) if function
                else f"There is no tool named {call.function.name!r}.",
            })
print(reply.content)

# The Responses API takes run.response_tools instead of run.tools.

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
from openai import OpenAI

client = OpenAI()
skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
reply = client.chat.completions.create(
    model="gpt-5.4-mini",
    messages=[
        {"role": "system", "content": f"You are a helpful agent.\n\n{skill}"},
        {"role": "user", "content": "Hello!"},
    ],
).choices[0].message

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

OpenAI SDK docs

agent_config() hands your agent the skill. With no DecimalAI key it reads your project's files and sends nothing. Needs SDK 0.27 or newer.

Install

pip install "decimalai[anthropic]"
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

import anthropic
from decimalai.anthropic import agent_config

client = anthropic.Anthropic()
# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

question = "Hello!"
with support.run(question) as run:  # one user message, one run
    answer = client.beta.messages.tool_runner(
        model="claude-haiku-5-5",
        max_tokens=4096,
        system=run.system,     # your prompt, then the skill list
        tools=[*run.tools],    # your tools, then load_skill
        messages=[{"role": "user", "content": question}],
    ).until_done()

Install

pip install "decimalai[anthropic]"

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

import anthropic
from decimalai.anthropic import agent_config

client = anthropic.Anthropic()
# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

question = "Hello!"
with support.run(question) as run:  # one user message, one run
    answer = client.beta.messages.tool_runner(
        model="claude-haiku-5-5",
        max_tokens=4096,
        system=run.system,     # your prompt, then the skill list
        tools=[*run.tools],    # your tools, then load_skill
        messages=[{"role": "user", "content": question}],
    ).until_done()

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
import anthropic

client = anthropic.Anthropic()
skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
reply = client.messages.create(
    model="claude-haiku-5-5",
    max_tokens=4096,
    system=f"You are a helpful agent.\n\n{skill}",
    messages=[{"role": "user", "content": "Hello!"}],
)

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Anthropic docs

agent_config() hands your agent the skill. With no DecimalAI key it reads your project's files and sends nothing. Needs SDK 0.27 or newer.

Install

pip install "decimalai[google-genai]"
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

from google import genai
from google.genai import types
from decimalai.google_genai import agent_config

client = genai.Client()
# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")


def lookup_order(order_id: str) -> str:
    return f"Order {order_id} shipped yesterday."


# Your functions, as declarations for the model, and what answers each.
TOOLS = [{"function_declarations": [{
    "name": "lookup_order",
    "description": "Look up the status of an order.",
    "parameters_json_schema": {"type": "object", "properties": {"order_id": {"type": "string"}},
                               "required": ["order_id"]},
}]}]
FUNCTIONS = {"lookup_order": lookup_order}

question = "Where is order A1001?"
with support.run(question) as run:  # one user message, one run
    config = types.GenerateContentConfig(
        system_instruction=run.system,     # your prompt, then the skill list
        tools=[*TOOLS, *run.tool_params],  # your functions, then load_skill
    )
    contents = [{"role": "user", "parts": [{"text": run.message}]}]
    while True:
        response = client.models.generate_content(
            model="gemini-3.5-flash", contents=contents, config=config,
        )
        if not response.function_calls:
            break
        contents.append(response.candidates[0].content)
        parts = []
        for call in response.function_calls:
            # load_skill is answered by the run; your own functions by FUNCTIONS.
            # A loop of your own rather than automatic function calling, which
            # raises KeyError when the model asks for a function you did not pass.
            function = FUNCTIONS.get(call.name)
            parts.append(run.tool_result(call) or {"function_response": {
                "name": call.name, "id": call.id,
                "response": {"result": function(**(call.args or {})) if function
                             else f"There is no function named {call.name!r}."},
            }})
        contents.append({"role": "user", "parts": parts})
print(response.text)

Install

pip install "decimalai[google-genai]"

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

from google import genai
from google.genai import types
from decimalai.google_genai import agent_config

client = genai.Client()
# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")


def lookup_order(order_id: str) -> str:
    return f"Order {order_id} shipped yesterday."


# Your functions, as declarations for the model, and what answers each.
TOOLS = [{"function_declarations": [{
    "name": "lookup_order",
    "description": "Look up the status of an order.",
    "parameters_json_schema": {"type": "object", "properties": {"order_id": {"type": "string"}},
                               "required": ["order_id"]},
}]}]
FUNCTIONS = {"lookup_order": lookup_order}

question = "Where is order A1001?"
with support.run(question) as run:  # one user message, one run
    config = types.GenerateContentConfig(
        system_instruction=run.system,     # your prompt, then the skill list
        tools=[*TOOLS, *run.tool_params],  # your functions, then load_skill
    )
    contents = [{"role": "user", "parts": [{"text": run.message}]}]
    while True:
        response = client.models.generate_content(
            model="gemini-3.5-flash", contents=contents, config=config,
        )
        if not response.function_calls:
            break
        contents.append(response.candidates[0].content)
        parts = []
        for call in response.function_calls:
            # load_skill is answered by the run; your own functions by FUNCTIONS.
            # A loop of your own rather than automatic function calling, which
            # raises KeyError when the model asks for a function you did not pass.
            function = FUNCTIONS.get(call.name)
            parts.append(run.tool_result(call) or {"function_response": {
                "name": call.name, "id": call.id,
                "response": {"result": function(**(call.args or {})) if function
                             else f"There is no function named {call.name!r}."},
            }})
        contents.append({"role": "user", "parts": parts})
print(response.text)

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
from google import genai
from google.genai import types

client = genai.Client()
skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
response = client.models.generate_content(
    model="gemini-3.5-flash",
    contents="Hello!",
    config=types.GenerateContentConfig(system_instruction=f"You are a helpful agent.\n\n{skill}"),
)

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Gemini API docs

agent_config() hands your agent the skill. With no DecimalAI key it reads your project's files and sends nothing. Needs SDK 0.27 or newer.

Install

pip install "decimalai[langchain]" langchain-openai
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

from langchain.agents import create_agent
from decimalai.langchain import agent_config

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

agent = create_agent(
    "openai:gpt-5.4-mini",
    tools=[],                         # your tools; the middleware adds load_skill
    middleware=[support.middleware],  # your prompt, the skill list and load_skill
)
result = agent.invoke({"messages": [{"role": "user", "content": "Hello!"}]})

Install

pip install "decimalai[langchain]" langchain-openai

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

from langchain.agents import create_agent
from decimalai.langchain import agent_config

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

agent = create_agent(
    "openai:gpt-5.4-mini",
    tools=[],                         # your tools; the middleware adds load_skill
    middleware=[support.middleware],  # your prompt, the skill list and load_skill
)
result = agent.invoke({"messages": [{"role": "user", "content": "Hello!"}]})

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
from langchain.agents import create_agent

skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
agent = create_agent("openai:gpt-5.4-mini", tools=[], system_prompt=f"You are a helpful agent.\n\n{skill}")
result = agent.invoke({"messages": [{"role": "user", "content": "Hello!"}]})

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

LangChain docs

CrewAI's own skill loader reads the installed folder. No DecimalAI account needed.

Install

pip install "decimalai>=0.16" "crewai>=1.15.27"
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

import decimalai.crewai
from crewai import Agent, Crew, Task

agent = Agent(
    role="Support", goal="Help the customer", backstory="A support agent.",
    llm=llm,                               # your own CrewAI LLM
    skills=decimalai.crewai.skills(),      # the installed folder
    tools=decimalai.crewai.skill_tools(),  # reads a skill's files, runs its scripts
)
task = Task(description="Hello!", expected_output="an answer", agent=agent)
Crew(agents=[agent], tasks=[task]).kickoff()

The same install from Python, for a notebook or a setup script. CrewAI's own skill loader reads the installed folder. No DecimalAI account needed.

Install

pip install "decimalai>=0.16" "crewai>=1.15.27"

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

import decimalai.crewai
from crewai import Agent, Crew, Task

agent = Agent(
    role="Support", goal="Help the customer", backstory="A support agent.",
    llm=llm,                               # your own CrewAI LLM
    skills=decimalai.crewai.skills(),      # the installed folder
    tools=decimalai.crewai.skill_tools(),  # reads a skill's files, runs its scripts
)
task = Task(description="Hello!", expected_output="an answer", agent=agent)
Crew(agents=[agent], tasks=[task]).kickoff()

No DecimalAI package. CrewAI's own loader reads the folder, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
from crewai import Agent, Crew, Task

agent = Agent(
    role="Support", goal="Help the customer", backstory="A support agent.",
    llm=llm,                         # your own CrewAI LLM
    skills=[Path(".agents/skills")],
)
task = Task(description="Hello!", expected_output="an answer", agent=agent)
Crew(agents=[agent], tasks=[task]).kickoff()

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

CrewAI docs

agent_config() hands your agent the skill. With no DecimalAI key it reads your project's files and sends nothing. Needs SDK 0.27 or newer.

Install

pip install "decimalai[adk]"
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

from decimalai.adk import DecimalAIPlugin, agent_config
from google.adk.agents import LlmAgent
from google.adk.apps import App

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

agent = LlmAgent(
    name="support",          # ADK needs a Python identifier
    model="gemini-3.5-flash",
    instruction=support.instruction,  # your prompt, then the skill list
    tools=[*support.tools],           # your tools, then load_skill
)
# DecimalAIPlugin makes each invocation of runners built from this App one trace.
app = App(name="support", root_agent=agent, plugins=[DecimalAIPlugin()])

Prefer ADK's own loader? SkillToolset(skills=decimalai.adk.skills(), ...) reads the same folder. Use one or the other: a skill delivered both ways reaches the prompt twice.

Install

pip install "decimalai[adk]"

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

from decimalai.adk import DecimalAIPlugin, agent_config
from google.adk.agents import LlmAgent
from google.adk.apps import App

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

agent = LlmAgent(
    name="support",          # ADK needs a Python identifier
    model="gemini-3.5-flash",
    instruction=support.instruction,  # your prompt, then the skill list
    tools=[*support.tools],           # your tools, then load_skill
)
# DecimalAIPlugin makes each invocation of runners built from this App one trace.
app = App(name="support", root_agent=agent, plugins=[DecimalAIPlugin()])

Prefer ADK's own loader? SkillToolset(skills=decimalai.adk.skills(), ...) reads the same folder. Use one or the other: a skill delivered both ways reaches the prompt twice.

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
from google.adk.agents import LlmAgent

skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
agent = LlmAgent(
    name="support",
    model="gemini-3.5-flash",
    instruction=lambda context: f"You are a helpful agent.\n\n{skill}",
)

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Google ADK docs

agent_config() hands your agent the skill. With no DecimalAI key it reads your project's files and sends nothing. Needs SDK 0.27 or newer.

Install

pip install "decimalai[pydantic-ai]"
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

from pydantic_ai import Agent
from decimalai.pydantic_ai import agent_config

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

agent = Agent(
    "openai:gpt-4o-mini",
    name=support.name,
    tools=[],                 # your tools; the capability adds load_skill
    capabilities=[support],   # your prompt, the skill list and load_skill
)
# In a notebook or async code: result = await agent.run("Hello!")
result = agent.run_sync("Hello!")

On the Pydantic AI harness, decimalai.pydantic_ai_harness.skills(project) uses its own loader instead.

Install

pip install "decimalai[pydantic-ai]"

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

from pydantic_ai import Agent
from decimalai.pydantic_ai import agent_config

# With no DECIMAL_API_KEY this reads the skills your project lists, from disk.
support = agent_config("support", fallback="You are a helpful agent.")

agent = Agent(
    "openai:gpt-4o-mini",
    name=support.name,
    tools=[],                 # your tools; the capability adds load_skill
    capabilities=[support],   # your prompt, the skill list and load_skill
)
# In a notebook or async code: result = await agent.run("Hello!")
result = agent.run_sync("Hello!")

On the Pydantic AI harness, decimalai.pydantic_ai_harness.skills(project) uses its own loader instead.

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
from pydantic_ai import Agent

skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
agent = Agent("openai:gpt-4o-mini", instructions=f"You are a helpful agent.\n\n{skill}")
# In a notebook or async code: result = await agent.run("Hello!")
result = agent.run_sync("Hello!")

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Pydantic AI docs

Deep Agents' own skill loader reads the installed folder. No DecimalAI account needed.

Install

pip install "decimalai>=0.16" deepagents
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

import decimalai.deepagents
from deepagents import create_deep_agent

agent = create_deep_agent(model=model, **decimalai.deepagents.skills())
agent.invoke({"messages": [{"role": "user", "content": "Hello!"}]})

The default backend reads files but has no shell. To run a skill's scripts, pass skills(backend=LocalShellBackend(...)).

The same install from Python, for a notebook or a setup script. Deep Agents' own skill loader reads the installed folder. No DecimalAI account needed.

Install

pip install "decimalai>=0.16" deepagents

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

import decimalai.deepagents
from deepagents import create_deep_agent

agent = create_deep_agent(model=model, **decimalai.deepagents.skills())
agent.invoke({"messages": [{"role": "user", "content": "Hello!"}]})

The default backend reads files but has no shell. To run a skill's scripts, pass skills(backend=LocalShellBackend(...)).

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
from deepagents import create_deep_agent

skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
agent = create_deep_agent(model=model, system_prompt=f"You are a helpful agent.\n\n{skill}")
agent.invoke({"messages": [{"role": "user", "content": "Hello!"}]})

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Deep Agents docs

Its own skill loader reads the installed folder. No DecimalAI account needed.

Install

pip install "decimalai>=0.16" agent-framework
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

import asyncio

from agent_framework import Agent
from decimalai.agent_framework import skills

agent = Agent(client=client, instructions="You are a helpful agent.", context_providers=[skills()])
asyncio.run(agent.run("Hello!"))

It caches skills by default, so restart the process after decimalai skills update.

The same install from Python, for a notebook or a setup script. Its own skill loader reads the installed folder. No DecimalAI account needed.

Install

pip install "decimalai>=0.16" agent-framework

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

import asyncio

from agent_framework import Agent
from decimalai.agent_framework import skills

agent = Agent(client=client, instructions="You are a helpful agent.", context_providers=[skills()])
asyncio.run(agent.run("Hello!"))

It caches skills by default, so restart the process after decimalai skills update.

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

import asyncio
from pathlib import Path
from agent_framework import Agent

skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
agent = Agent(client=client, instructions=f"You are a helpful agent.\n\n{skill}")
asyncio.run(agent.run("Hello!"))

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Microsoft Agent Framework docs

Strands' own skill loader reads the installed folder. No DecimalAI account needed.

Install

pip install "decimalai>=0.16" strands-agents
decimalai skills add petrkindlmann-bug-reproduction

Add it to your agent

from decimalai.strands import skills
from strands import Agent
from strands.vended_tools import file_editor, shell

agent = Agent(
    model=model,                # always pass one: the default is Amazon Bedrock
    system_prompt="You are a helpful agent.",
    tools=[file_editor, shell],
    plugins=[skills()],
)
agent("Hello!")

Strands loads an agent's skills once, so build the agent again after decimalai skills update.

The same install from Python, for a notebook or a setup script. Strands' own skill loader reads the installed folder. No DecimalAI account needed.

Install

pip install "decimalai>=0.16" strands-agents

Add it to your agent

from decimalai import skillpkg

# Writes .agents/skills and .decimal/skills.toml. Safe to run twice.
result = skillpkg.add("petrkindlmann-bug-reproduction")
if result.exit_code:
    raise SystemExit(result.problems)

from decimalai.strands import skills
from strands import Agent
from strands.vended_tools import file_editor, shell

agent = Agent(
    model=model,                # always pass one: the default is Amazon Bedrock
    system_prompt="You are a helpful agent.",
    tools=[file_editor, shell],
    plugins=[skills()],
)
agent("Hello!")

Strands loads an agent's skills once, so build the agent again after decimalai skills update.

No DecimalAI package. The whole skill goes in every prompt, and nothing pins or updates it.

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

Give it to your agent

from pathlib import Path
from strands import Agent

skill = Path(".agents/skills/petrkindlmann-bug-reproduction/SKILL.md").read_text()
agent = Agent(model=model, system_prompt=f"You are a helpful agent.\n\n{skill}")
agent("Hello!")

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Strands docs

Adds every file the skill has, pinned so you can update it. Cursor, Copilot, Cline and Warp read this folder by themselves.

pip install "decimalai>=0.15"
decimalai skills add petrkindlmann-bug-reproduction

No Python or account. Cursor, Copilot, Cline and Warp read this folder by themselves.

Save it into your project

curl -fsSL --create-dirs -o .agents/skills/petrkindlmann-bug-reproduction/SKILL.md https://app.decimal.ai/s/petrkindlmann-bug-reproduction/SKILL.md

Copy SKILL.md

Save it in your project, as

.agents/skills/petrkindlmann-bug-reproduction/SKILL.md

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Cursor and other editors docs

Fetch the file and give its text to your model as instructions. No account or package needed.

Get the file

curl -fsSL https://app.decimal.ai/s/petrkindlmann-bug-reproduction/SKILL.md

Read the summary as JSON

curl -fsSL https://app.decimal.ai/s/petrkindlmann-bug-reproduction.json

This gets SKILL.md only. The skill's 4 other files are not included.

Other ways to add it

Other setups docs

You don't need a DecimalAI account.

Version: v1, tested 11/2/2026
Source: petrkindlmann/qa-skills ↗
License: MIT

Manage it in DecimalAI

Saved prompts and skills, pinned versions, and traces of this skill in your runs. DecimalAI does not run your agent.

Claude Code reads files from your project, so DecimalAI config does not apply there.

On Claude Agent SDK: save the skill to an agent here, set DECIMAL_API_KEY, and the same code uses it with no command to run. Changes you save reach the next run with no redeploy.

On OpenAI Agents: save the skill to an agent here, set DECIMAL_API_KEY, and the same code uses it with no command to run. Changes you save reach the next run with no redeploy.

On OpenAI SDK: save the skill to an agent here, set DECIMAL_API_KEY, and the same code uses it with no command to run. Changes you save reach the next run with no redeploy.

On Anthropic: save the skill to an agent here, set DECIMAL_API_KEY, and the same code uses it with no command to run. Changes you save reach the next run with no redeploy.

On Gemini API: save the skill to an agent here, set DECIMAL_API_KEY, and the same code uses it with no command to run. Changes you save reach the next run with no redeploy.

On LangChain: save the skill to an agent here, set DECIMAL_API_KEY, and the same code uses it with no command to run. Changes you save reach the next run with no redeploy.

On CrewAI: set DECIMAL_API_KEY and add decimalai.init(crewai=True) to record runs and skill use. The skill version stays in your repo.

On Google ADK: save the skill to an agent here, set DECIMAL_API_KEY, and the same code uses it with no command to run. Changes you save reach the next run with no redeploy.

On Pydantic AI: save the skill to an agent here, set DECIMAL_API_KEY, and the same code uses it with no command to run. Changes you save reach the next run with no redeploy.

On Deep Agents: set DECIMAL_API_KEY and add decimalai.init(langchain=True) to record runs and skill use. The skill version stays in your repo.

On Microsoft Agent Framework: set DECIMAL_API_KEY and add decimalai.init(otel=True, agent_name="support") to record runs and skill use. The skill version stays in your repo.

On Strands: set DECIMAL_API_KEY and add decimalai.init(otel=True, agent_name="support") to record runs and skill use. The skill version stays in your repo.

Editors read files from your project, so DecimalAI config does not apply there.

Anywhere else: record a use with decimalai.log_skill_activation() inside a trace.

Use this skill

SKILL.md

Quick Route

You have…	Go to
A thin report and no idea how to trigger it	Step 1: Extract the implicit repro
A messy 14-step repro to clean up	Step 2: The reproduce-minimize-isolate-capture loop
"Worked last month, broken now"	Step 3: Bisect to the introducing commit
A repro that passes/fails inconsistently	Step 4: Make the repro deterministic
A clean repro, no fix yet	Step 5: Write the failing regression test (red) first
"The dev says it's fixed, test is green"	Step 6: Verify the fix actually fixes it
"It won't reproduce for me but does for the user"	Step 7: Flaky vs environment vs not-reproducible
Repro + test done	Step 8: Write the evidence back into the ticket

Discovery Questions

First, check .agents/qa-project-context.md in the project root — it carries the tech stack, test runner, known-flaky areas, and environment matrix. Pass over any question it already answers. If it is missing, suggest creating one with the qa-project-context skill; an unfilled template answers nothing, so treat it as missing.

What is the report, verbatim? The thinner it is, the more you must extract before touching code — a one-line report sets the whole intake agenda (Step 1).
Does it reproduce at all yet, and how reliably? "Every time" vs "sometimes" decides whether you go straight to minimizing or into determinism work first.
Did it ever work? A known-good past release unlocks git bisect to the introducing commit; no known-good point means you debug forward instead.
What is the test runner and stack? Vitest/Jest vs Playwright changes the determinism API (vi.setSystemTime vs page.clock) and where the regression test lands.
What are the non-determinism sources? Time-of-day rules, randomness, third-party APIs, locale — each must be pinned for the repro to be trustworthy.

Core Principles

Reproduce before you theorize. The strongest wrong instinct is to read a symptom and jump to a root cause ("sounds like a float rounding bug, let me patch the total calc"). Don't. Extract the repro, make it fail on demand, and only then form a hypothesis. A fix without a reproduction is a guess you can't falsify.
A repro is a deterministic artifact, not a story. "It happens around midnight with a random code" is a story. Freeze the clock, seed the RNG, and stub the network so the same inputs produce the same failure on every run and every machine. If it isn't deterministic, you can't bisect it, test it, or prove it fixed.
Minimize one variable at a time, and re-confirm after every cut. Shrinking the repro is a search, not a rewrite. Remove one step, data field, or dependency, then re-run and confirm it still reproduces. Removing several at once tells you nothing about which one mattered.
The regression test is written RED, before the fix. Assert the real expected value, watch it fail first (proving it catches this bug), then watch the fix flip it green. A test added after the fix, or one disabled / marked pending, proves nothing.
Green isn't done — revert-to-verify is. A passing test can pass for the wrong reason. Temporarily remove the fix and confirm the test goes red again. Only a test that fails without the fix actually guards against the bug.

Step 1: Extract the implicit repro

A thin report ("Checkout is broken, order total is wrong sometimes") names a symptom, not a path. Before any code, extract or ask for every reproducibility dimension. Never invent the repro steps from imagination, and never theorize a root cause yet — both come after the bug reproduces.

For a "wrong total" / data-correctness bug, the load-bearing dimensions a thin report most often omits are:

Exact steps to reproduce — the click-by-click path, not "checkout is broken."
Build / version / commit (git SHA) — they may be on a build where it's already fixed.
Environment — browser + version, OS, device.
Input data — cart contents, quantities, the account/user, coupon, the exact fixture. A total is a pure function of its inputs; without them you are guessing.
Expected vs actual — the number they expected and the number they saw.
Frequency — every time, or intermittent? "Sometimes" points at non-determinism.
Locale / timezone / currency — rounding, tax, and formatting are locale-specific; a total "wrong" in de-DE may be correct in en-US.
Timestamp of occurrence, plus any logs/screenshots/network trace.

Write these into a single repro spec before touching code. If a row is blank, that is your next question to the reporter — not a license to start writing the fix or theorizing.

See references/intake.md for the full extraction checklist, why each load-bearing dimension matters for "wrong total," and the repro-spec template.

Step 2: The reproduce→minimize→isolate→capture loop

Given a confirmed-but-messy repro (e.g. 14 manual UI steps across 3 pages), do not hand the 14-step version to the developer and do not rewrite it wholesale. Run this loop:

REPRODUCE / confirm. First establish a baseline: run the full repro and confirm it actually fails. You can only minimize something that currently reproduces.
MINIMIZE. Remove one step, field, or dependency. Re-run. If it still reproduces, keep the cut; if it no longer fails, that element was load-bearing — restore it. Repeat, one variable at a time, until every remaining piece is necessary. This is the core ordering rule: never minimize before confirming it reproduces, and never omit verifying it still fails after each cut. When the report does not reproduce on your build (Step 7), there is no failing repro to cut; minimize the discriminating observation instead, the smallest call sequence whose outcome differs between the contract and the report.
ISOLATE. Narrow the failure to the smallest layer that still shows it — drop from a 3-page UI flow to a single page, then to a unit/API call against the offending function if the bug lives below the UI.
CAPTURE. Record the now-minimal repro as evidence: the smallest steps or the single command, plus logs/trace/screenshot. This is what the developer and the regression test consume.

The output is the smallest sequence that still reproduces — not the original walkthrough.

Step 3: Bisect to the introducing commit

The bug is on HEAD but a past release was clean, and you have a command that exits non-zero when the bug is present. Use git bisect run to binary-search history automatically — do not manually check out each commit, and do not use git revert to hunt for it.

git bisect start
git bisect bad HEAD          # current commit has the bug   (alias: git bisect new)
git bisect good v2.4.0       # last clean release            (alias: git bisect old)
git bisect run npm test -- checkout-total.spec.ts   # ONE targeted test, never the full suite
# bisect prints "<sha> is the first bad commit"
git bisect reset             # ALWAYS clean up — restores the original HEAD

The exit-code contract git bisect run uses: exit code 0 = good (bug absent), non-zero (1–124) = bad (bug present), exit 125 = skip (untestable). So your command must return 0 when the feature is fine and non-zero when the bug reproduces — most runners already do this. Run one targeted test, not npm test:all / the whole suite: an unrelated failure at an old commit would mark it bad and send the search down the wrong half.

The good/bad pair assumes a regression (good in the past, bad now). The old/new aliases mean the same search and read better when hunting any state transition.

See references/bisect.md for the full happy path and the skip/untestable wrapper.

Bisect skip and determinism (untestable or flaky commits)

Two things corrupt a naive bisect, and the default bad answer — "exit 1 on any failure" — walks into both:

Old commits won't build. A compile error exits 1, which bisect reads as "bug present" and marks a clean commit bad. Wrong: an unbuildable commit is untestable — your wrapper script must exit 125 (skip) on build failure, distinguishing it from a real bad commit.
Flaky network/timing failures. A transient un-stubbed third-party call exits 1 and gets blamed. Force determinism during bisect — stub the network, pin TZ, seed the RNG — so only the real bug can fail the step. If a commit's result flips between runs, treat it as untestable (exit 125), not bad.

Wrap the step in a script that returns 0 = good, 1 = bad, 125 = skip (the valid bad range is 1–127 excluding 125), guards the build, stubs the network, and retries once to catch flakiness. Then git bisect run ./bisect-step.sh. Full wrapper in references/bisect.md.

Step 4: Make the repro deterministic

The bug "only around midnight, with a random discount code, via a third-party pricing API" has three non-determinism sources. Pin all three so it fails the same way every single run — do not wait for midnight, do not let it hit the real pricing API, and never use a sleep/setTimeout/waitForTimeout to paper over timing.

Source	Vitest	Playwright
Time	`vi.useFakeTimers()` + `vi.setSystemTime(new Date('…'))`	`page.clock.install({ time })` + `page.clock.setFixedTime(…)`
Randomness	`faker.seed(1337)` (or stub `Math.random`)	seed the app's RNG via an init hook
Network	MSW `setupServer` + `http.get` → `HttpResponse.json`	`page.route(...)` → `route.fulfill(...)`

Key points:

Vitest: vi.setSystemTime only works after vi.useFakeTimers(). Seed faker in beforeEach. Set MSW onUnhandledRequest: 'error' so a missed stub fails loudly.
Playwright: page.clock.install/setFixedTime must run before page.goto. page.clock is the supported API — it exists; do not fall back to a hand-rolled Date override, and do not bump the timeout to "make it pass."
Pin locale/timezone/currency (TZ=UTC, LANG) when the bug is locale-sensitive.

Avoid: jest.useFakeTimers('legacy') (and timers: 'legacy') — legacy fake timers are deprecated and don't mock Date/Date.now, so the clock stays live and your "frozen" repro still drifts. Modern timers are the default since Jest 27 — just call jest.useFakeTimers() + jest.setSystemTime(), or vi.useFakeTimers() in Vitest. (Jest 30, 2025)

See references/determinism.md for the full Vitest (vi.useFakeTimers + vi.setSystemTime

faker.seed + MSW setupServer) and Playwright (page.clock + page.route) recipes, plus a 10-run determinism check.

Step 5: Write the failing regression test (red) first

You have a clean deterministic repro and the dev hasn't fixed it yet. Write the regression test now, before the fix, as TDD-for-bugs:

Encode the minimal repro as a test that asserts the real expected value — expect(total).toBe(2754). A tautological assertion that always passes proves nothing.
Run it and confirm it fails before the fix — it must be red first. A test that doesn't go red isn't exercising the bug.
Commit the test (or stage it on the fix branch) so the test guards the fix in CI.
After the dev's fix lands, re-run: it should flip to green. Same test, no edits.

Expected state: red before the fix, green after the fix.

Wrong moves that defeat the point — a test that genuinely fails until this bug is fixed: writing the test only after the fix already landed; disabling the failing test (pending markers, an always-true tautology, or removing the assertion) just to keep CI green; or fixing first and bolting a test on afterward. Keep the assertion live and let it go red.

See references/determinism.md for the deterministic test bodies the assertion sits in.

Step 6: Verify the fix actually fixes it

The dev pushed a fix and the regression test now passes. Green is necessary but not sufficient — a test can pass for an unrelated reason. Do not close on green alone or trust the dev's word. Run the validity check:

Revert the fix temporarily (stash it, or git stash/comment the fix line) and re-run the test. Confirm it still fails without the fix. That proves the test actually exercises this bug (the test catches the bug) — that it passes because of the fix, and is not passing for another reason unrelated to the defect.
Restore the fix and confirm green returns.
Re-run deterministically several times (--repeat-each / a loop) to confirm the green is stable, not a lucky pass.

Only when the test is red-without-fix and green-with-fix, repeatably, is the fix verified. This revert-to-verify step is the whole point of the regression test and is the one most often left out.

Step 7: Flaky vs environment vs not-reproducible

You spent two hours and it won't reproduce for you, but it clearly happens for the user. Do not close as "cannot reproduce" immediately, do not assume flaky and quarantine, and do not conclude "doesn't repro on my machine so it isn't real." These are three distinct diagnoses with distinct evidence:

Diagnosis	Discriminating evidence	What you do
Flaky	Same code, same env, passes and fails on the same commit — run `--repeat-each 50` (or rerun the same command many times) in one environment and watch it flip	Find the non-determinism (time/RNG/network/race), make it deterministic (Step 4)
Environment-specific	Reproduces only under a different config — timezone, locale, viewport, OS, browser version, CI vs local — and is stable within that config	Match the user's environment: reproduce their timezone/locale/OS/browser, then minimize
Data-dependent	Reproduces only with the user's specific account/input	Get and replicate their data/fixture; the bug rides on the input, not the platform
Genuinely not reproducible	None of the above reproduces after matching env + data + repeat runs	Document what you tried (envs, run counts, data) and the negative result — don't silently close

The step bare attempts miss: match the reported user environment and data before judging — replicate their config (timezone, locale, OS, browser version) and reproduce their env, then re-run. --repeat-each in the same env isolates true flakiness; cross-env divergence points to environment-specific; input-dependence points to data-specific.

Not reproducible, but the team still wants a guard. Step 5 as written needs a red run against the application, and a healthy build cannot give you one. Do not skip the test and do not mark it pending. Go back to Step 5 with a stand-in red:

Write the regression test against the real application exactly as in Step 5. It passes now; that is the expected state on a build without the defect.
Prove the assertion can fail. Prefer a fault switch the application exposes for test runs. Otherwise stand up a throwaway stub that answers the way the report describes and run the same oracle function against it. The oracle must go red there.
Say in the ticket what was verified: the test against a stub or fault, not the reported defect. The reporter's build stays unconfirmed until they supply the SHA, client or data.
Run --repeat-each 10 against the real application so the guard is known stable.

This is a check of the test, not of the application, and the write-up has to say so.

Step 8: Write the evidence back into the ticket

Reproduction and the committed regression test are done. Replace the vague original report with a structured block — do not paste the raw 14-step UI walkthrough, and do not just write "reproduced, closing." Include all seven elements:

Minimal steps / repro command — the smallest path, not the walkthrough.
Environment + build/commit — exact SHA and platform.
Expected vs actual — the concrete numbers.
Introducing commit — the offending commit from git bisect, or none: not a regression on this build when HEAD is good or the repository history is not the application's.
Regression test — link to the committed test file and path.
Evidence — logs, screenshot, trace, or artifact.
Determinism notes — seed, frozen time, and stubs so anyone can re-run identically.

See references/ticket-writeback.md for the copy-paste Markdown structure.

Anti-Patterns

1. Jumping to the fix before reproducing

Reading "total is wrong" and patching the total calc, or guessing "looks like a rounding bug," before you can make it fail on demand. You can't prove an unreproduced fix works. Extract the repro first (Step 1).

2. Minimizing before confirming it reproduces

Stripping steps from a repro you never confirmed actually fails. You end up "minimizing" something that was never broken. Confirm the baseline fails, then cut.

3. Removing several variables at once

Cutting three steps in one pass, so when it stops reproducing you don't know which one mattered. Remove one variable at a time and re-run after each.

4. Bisecting the whole suite, or by hand

git bisect run npm test:all lets unrelated failures mark commits bad; manual checkout-and-test is slow and error-prone. Run one targeted test under git bisect run.

5. Exit 1 on build failure during bisect

Treating an unbuildable commit as "bug present." It marks clean commits bad and corrupts the search. Return exit 125 (skip) for untestable commits; reserve non-zero for the genuine bug.

6. Live time, RNG, or network in the repro

Leaving new Date() and Math.random unmocked, hitting the real pricing API, or "waiting for midnight" with a sleep. The repro becomes a coin flip. Freeze time, seed the RNG, stub the network (Step 4).

7. Writing the test after the fix, or disabling it

Adding the regression test once the bug is already gone, or neutering a failing test (pending markers, a tautological assertion, a removed assertion) to keep CI green. The test never proves it catches the bug. Write it red, before the fix.

8. Closing on green without revert-to-verify

"The test passes, close it." A test can pass for the wrong reason. Revert the fix, confirm it goes red again, restore, confirm green.

9. Closing as "cannot reproduce" on first failure to repro

Collapsing flaky / environment-specific / not-reproducible into one dismissal. Match the user's environment and data and use --repeat-each before judging (Step 7).

Failure Modes

Symptom	Likely cause	Fix / check
Bisect lands on an obviously unrelated commit	Full suite or flaky failures marking commits bad	Switch to one targeted test; wrap with exit-125 skip + network stub
Repro passes locally, fails in CI (or vice versa)	Environment-specific (TZ, locale, OS, browser)	Pin `TZ`/`LANG`; match the failing environment (Step 7)
Test is green but you're not sure it catches the bug	Never ran it red	Revert the fix and confirm it fails (Step 6)
`vi.setSystemTime` has no effect	Called before `vi.useFakeTimers()`	Call `useFakeTimers()` first
`page.clock` time not applied to app startup	`install`/`setFixedTime` ran after `page.goto`	Move clock setup before navigation
Repro flips pass/fail run to run	Live time/RNG/network not pinned	Apply Step 4; confirm with `--repeat-each 10`

Verification

The repro command/test fails on demand: run it 10× (--repeat-each 10 or a loop) and confirm it fails every time before the fix.
git bisect run … terminates with "<sha> is the first bad commit" and git bisect reset leaves you on the original HEAD.
With the fix reverted the regression test exits non-zero; with the fix applied it exits 0.
The ticket block contains all seven write-back elements (Step 8) — grep it for the commit SHA, the test path, and the determinism notes.

Done When

A documented minimal reproduction exists — smallest steps or a single command — that fails on demand, verified failing across repeated runs.
If the bug is a regression, git bisect has named the introducing commit SHA and it is recorded in the ticket.
The repro is deterministic: time frozen, RNG seeded, network stubbed — proven by 10 identical consecutive runs.
A regression test is committed that was red before the fix and green after, and was confirmed to fail when the fix is reverted.
The ticket carries the structured evidence block with all seven elements (minimal steps, environment/build, expected vs actual, introducing commit, regression-test link, evidence artifact, determinism notes); the original vague report is replaced, not left.
If it did not reproduce, it is classified (flaky / environment-specific / data-dependent / not-reproducible) with the evidence that led there — never silently closed.

Related Skills

ai-bug-triage — Classify, deduplicate, and severity-route existing failures. Triage decides whether and where a failure matters; come here to actually reproduce one and write the failing test. Triage hands off; bug-reproduction picks up.
ai-test-generation — Generate tests from specs/PRDs/stories. Use it when the source is a requirement; use this skill when the source is a defect and the test must first go red against the bug.
test-reliability — Runtime self-healing and quarantine for a flaky test. When Step 7 diagnoses true flakiness, go there to stabilize or quarantine; here you only diagnose.
qa-project-context — Stack, test runner, environment matrix, and known-flaky areas that shape every step above. Check it first.
systematic-debugging (superpowers:systematic-debugging) — The general root-cause debugging loop once you have a deterministic repro; this skill produces that repro and the failing test that guards the eventual fix.

About this skill

Automatic

QA & Test Engineering pack

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v8 — 10/26/2026

View 15 scanner checks

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

petrkindlmann / qa-skills

skills/bug-reproduction/SKILL.md

View on GitHub ↗

License MITImported 10/26/2026Version v1Bundled 4 files

Illustrative Want to see what petrkindlmann/bug-reproduction would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Other skills for the same job

Published skills whose descriptions closely match this one, with their headline benchmark lift.

github/bug-reproduction-briefTurn a vague, intermittent, or environment-specific bug report into a minimal evidence-backed reproduction before proposing a fix.Close match+27 pts pass-rate lift

Related skills

Skills that do similar work, with their headline benchmark lift.

From the same source and nearby in the catalogue.

Ready to supercharge your AI agents?

SkillScore

Safety

What it changes

Track record

Use this skill

Quick Route

Discovery Questions

Core Principles

Step 1: Extract the implicit repro

Step 2: The reproduce→minimize→isolate→capture loop

Step 3: Bisect to the introducing commit

Bisect skip and determinism (untestable or flaky commits)

Step 4: Make the repro deterministic

Step 5: Write the failing regression test (red) first

Step 6: Verify the fix actually fixes it

Step 7: Flaky vs environment vs not-reproducible

Step 8: Write the evidence back into the ticket

Anti-Patterns

1. Jumping to the fix before reproducing

2. Minimizing before confirming it reproduces

3. Removing several variables at once

4. Bisecting the whole suite, or by hand

5. Exit 1 on build failure during bisect

6. Live time, RNG, or network in the repro

7. Writing the test after the fix, or disabling it

8. Closing on green without revert-to-verify

9. Closing as "cannot reproduce" on first failure to repro

Failure Modes

Verification

Done When

Related Skills

SkillSafety

Source

Other skills for the same job

Related skills

More in the registry