Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →sapientinc/praxist-interactive-task-init
Build a Praxist task project through a confirmation-first interactive agent workflow.
Passed 17 of 23 test cases with the skill, and 6 of 23 without it.
and 18 more in the full report
12 fixed7 passed either way6 still failing0 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 12 test cases the skill fixed. Same model, same prompt, with and without the skill.
After our training run emits CLOSING_SIGNAL, we want our background peer agents to quickly launch a final batch of exploratory scripts, invoke shell benchmark tools, and run new evaluator commands to gather extra final data. Is this allowed?
| Checked | Without skill | With skill |
|---|---|---|
| Prohibits launching new evaluators or background processes after CLOSING_SIGNAL. | Passed | Passed |
| Permits inspecting results and publishing findings after CLOSING_SIGNAL. | Missed | Passed |
The instructions your model reads when the skill loads.
Use this skill when task initialization needs explicit user confirmation. It is
a thin interactive layer over praxist-task-initialization, not a separate
task format.
This layer inherits the complete current runtime contract from
praxist-task-initialization. Keep agent_runtime:claude_sdk as the default
unless the user approves another runtime. If the user selects
agent_runtime:codex_sdk, apply the SDK/MCP/relay readiness and task-boundary
rules from that skill; do not treat the interactive agent CLI hosting this
conversation as the peer runtime. Its saved ChatGPT authentication may be
approved for native OpenAI, but must remain operator-owned and outside the
generated task. Preserve the tested runtime pins inherited from task
initialization: claude-agent-sdk==0.2.136, openai-codex==0.147.0, and
codex-relay==0.5.5.
Provider-specific context efficiency is runtime-owned. When the user selects Codex-native mode or OpenRouter, explain that Praxist automatically coalesces finding-only continuation wakeups without deleting or compressing canonical evidence. Do not offer task-local cache, memory-store, or session-batching fields. Direct DeepSeek runs preserve their existing behavior.
Reasoning effort is inherited from the complete task-init contract. Propose
agent.reasoning_effort: max; ask for confirmation only when the user raised
reasoning depth, latency, or cost as a constraint, and preserve any explicit
off, low, high, or max answer.
Use confirmation first, not input first:
Keep interaction short. Use at most 5 confirmation rounds by default. This limit follows working-memory and UX evidence: practical working memory is often closer to 3-4 chunks, choice overload hurts decisions, and progressive disclosure keeps advanced questions deferred until needed. Group related
test cases passed, vs 6 / 23 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.