delegate-analysis
Skill60 starsChanged 34 days agoMITView on GitHub
Skill-driven exploration, targeted reading, web research, and deep-dive analysis for delegate agents. Use when the primary delegates investigation or information gathering rather than artifact review or code execution.
In this repository17 other files it gives its agents.
- AGENTS.mdAGENTS.md
- docs/agents.mdAGENTS.md
- archive-legacy-docsSkill
- author-and-verify-implementation-planSkill
- browser-walkthroughSkill
- create-planSkill
- execute-work-packageSkill
- generate-docsSkill
- generate-handoverSkill
- report-environment-issueSkill
- resume-planSkill
- review-fixSkill
- review-implementation-planSkill
- review-implementationSkill
- review-planSkill
- update-docsSkill
- update-planSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 13
Use one of the following modes.
- Distinctive: line 25
Use the named files as scoped starting points and retrieve only what answers the question.
- Distinctive: line 31
Prefer authoritative primary sources.
- Distinctive: line 38
Start at the named entry point and follow relevant indirections until the question is answered.
- Very specific: line 44
Default to read/analyze/verify.
- Distinctive: line 49
If the task becomes a review, artifact authoring task, or implementation task, stop and ask the primary to route through the matching skill.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Skill: Delegate Analysis
Use one of the following modes. The primary should name the mode, scope, and question explicitly.
Modes
code-exploration
- Orient with file search, then read only the key files.
- Trace dependencies one level further only when needed.
- Return findings with concrete file paths and symbols.
targeted-reading
- Use the named files as scoped starting points and retrieve only what answers the question.
- Extract only the facts needed to answer the question.
- Organize the result by topic or file, with line references where useful.
web-research
- Prefer authoritative primary sources.
- Cross-check important claims when practical.
- Return a concise synthesis with source URLs and explicit uncertainty.
- This is delegate work: use
retrieveronly to crawl or fetch already-selected sources, not for open-ended search, source selection, or cross-source judgment.
deep-dive
- Start at the named entry point and follow relevant indirections until the question is answered.
- Map the code or data path with file and symbol evidence.
- Highlight non-obvious behavior, assumptions, and unresolved gaps.
Rules
- Default to read/analyze/verify. Do not edit files.
- Stay within the delegated scope.
- Directly read scoped source, docs/plans, symbols, and compact targeted evidence. Keep uncurated bulk evidence out of the delegate context: use a reliable focused filter when sufficient; otherwise route the raw artifact, command, or path plus a focused question to
retriever, including coherent multi-file evidence. Numeric tool truncation is a safety net, not the routing rule. - Spool potentially verbose command output to a predictable path under
/tmp/opencode/; retain only path, command, exit status, and compact metadata/evidence. This supports same-machine continuation after an interruption, not reboot durability. - The parent owns synthesis and verifies only evidence that materially supports its conclusions rather than repeating broad retrieval.
- If the task becomes a review, artifact authoring task, or implementation task, stop and ask the primary to route through the matching skill.
- After an analysis digest/message diagnoses an evidence-backed defect, the Primary may accept it and explicitly instruct remediation by resuming this session through
review-fix. That skill owns eligibility, writes, and the one-pass stop; analysis itself remains read-only. Reuse correct and sufficient diagnosis context to avoid collecting secured evidence again; otherwise follow its fresh-session fallback with a finding-source hint. Conversation-owned plan corrections still route toupdate-plan. - Return a compact digest, not raw file contents or long logs.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

