Gather task-agnostic scientific research context for Praxist task projects using no-key public literature/database/open-access lookup, agent-host web search when available, local project documents, and source/provenance notes.
SkillScore78Maintained
Automatic
◆Research pack
Last verified 9/28/2026
What it changes
Passed 15 of 19 test cases with the skill, and 3 of 19 without it.
Without skill17%
With skill78%
Checks it passed that the model alone missed
rejects requiring additional paid API keys
groups research directions by intervention surface
documents missing resources as local adaptation notes
and 15 more in the full report
14 fixed4 passed either way5 still failing0 broke
Pass-rate lift
+61 pp
Tokens
+6%
Agent turns
0%
Response time
-46%
Model gemini-3.6-flashMeasured 9/28/2026v118 of 23 cases passed with skill · 19 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 14 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 14
Prompt
We need to check the exact methodology section of an open-access arXiv paper (arXiv:2301.00001) for hyperparameter settings. Which tool function extracts open-access text up to a maximum character limit?
What was checked, without and with the skill
Checked
Without skill
With skill
identifies literature_open_access_text as the tool function for extracting open-access full text
Missed
Passed
case-13 · gemini-3.6-flash · 9/28/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md100 linessingle file
Praxist Scientific Research
Use this skill to build a source-backed research context for a Praxist task
project. This is a planning and context-gathering skill, not an experiment
runner.
Boundaries
Do not start, stop, resume, or mutate a Praxist run.
Do not edit Praxist core or generic plugins.
Do not require additional API keys. Prefer public no-key sources and Codex web
search when available.
Do not treat papers, database entries, blog posts, benchmark pages, or
open-access text as measured task performance.
Do not download datasets, checkpoints, simulators, packages, licensed assets,
or new runtime environments just because a source mentions them. Use those
sources to improve the best solution possible under the task's current local
data, simulator, dependencies, evaluator, hardware, and runtime. Record
missing resources only as task-local notes or user-facing requirements.
Keep domain-specific search policy and conclusions in the task project,
usually under assets/literature/, description.md, role skills, or task
prompt files.
Source Priority
Local project docs, papers, README files, benchmark docs, existing logs, and
task-owned assets/literature/.
Public no-key literature/database tools when tool_server:literature_lookup
is available in the standard or task-declared Praxist tool set.
Agent-host web search/open web browsing for official benchmark pages, dataset
cards, standards, methods, and recent papers.
Optional credentialed or licensed sources only when the user explicitly
provides them and the task requires them.
Workflow
Identify the task or research project root. Prefer the current directory when
the user does not provide a path.
Draw readable ASCII line charts directly in the terminal from numeric series, command output, CSV, JSON, or manually extracted points. Use when the user asks for a curve, trend line, score progression, leaderboard trend, metric history, or any plot that should be visible in a CLI/chat transcript without opening a GUI or writing image files.