Claude Skill

research

Investigate a question against high-trust primary sources and capture the findings as a Markdown file in the repo. Use when the user wants a topic researched, docs or API facts gathered, or reading legwork delegated to a background agent.

LLM Mart · 0 points · 15 views 0 listing impressions 0 install-command copies
Virus-scanned Reviewed automatically before listing.

Full trust report

Download ConnorGriffin-skills-skills_tools_research-afea794.zip · 1 KB
Part of connorgriffin/skills — 25 skills

Install

skills CLI npx skills add https://github.com/ConnorGriffin/skills/tree/main/skills/tools/research
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install connorgriffin-skills@llmmart
Git git clone https://github.com/ConnorGriffin/skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole connorgriffin/skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Research-worker dispatch

Use exactly one research worker; never create a chain of workers.

The coordinator supplies the selected adapter, explicit research-worker model, explicit research-worker effort, and the complete research task. Dispatch only through skills/drivers/orchestrate/scripts/codex-worker.py or skills/drivers/orchestrate/scripts/claude-worker.py, using the selected adapter's read-only surface. Never use the built-in Agent tool, Workflow tool, or background-agent machinery.

Use the provider-neutral hosted-source capability defined by the selected adapter's dispatch reference on every read-only research start, with one coordinator-owned state file under the coordinator's session-scratch directory. The selected adapter's resume, stop, and verify surfaces retain lifecycle ownership. Do not restate provider argv here.

After selection, this interface does not reclassify research work or choose a model or effort. Preserve adapter-owned state and coordinator-owned recovery through the orchestrate adapter contract; do not restate its command or lifecycle mechanics here.

Before dispatch, the coordinator writes the exact complete research-task prompt bytes to an immutable session-scratch file and passes that file's contents as the selected adapter's positional prompt. The worker receives no chat transcript or other coordinator-session material. The prompt requires the worker to return a final message beginning SOURCE_ACCESS_UNAVAILABLE: when hosted search or fetch is refused, and to write no findings document in that case.

Only when that start completed with a terminal, resumable session and its final message begins with the exact SOURCE_ACCESS_UNAVAILABLE: prefix may the coordinator fetch the required public sources. Create a unique .research-sources.* directory under the worker cwd, store only fetched public unauthenticated source files plus manifest.md mapping each file to its source URL, and resume the same worker with the manifest's absolute path. This bounded handoff is the sole narrow exception to the original-prompt-only rule and takes precedence over it. The directory is removed after completion.

If the adapter leaves no resumable session or the coordinator cannot obtain the required primary sources, stop explicitly and write no successful findings document. Never put transcripts, credentials, secrets, .env, patient data, authenticated or private source material, or unrelated coordinator-session content in the handoff.

The worker performs the research directly and returns source-cited findings to the coordinator. Never spawn another background agent or nested worker. If the worker fails or is interrupted, report the failure explicitly. Do not describe a successful dispatch as completed research. The coordinator writes the returned findings to the single Markdown file required below.

The research worker's job

  1. Investigate the question against primary sources — official docs, source code, specs, first-party APIs — not a secondary write-up of them. Follow every claim back to the source that owns it.
  2. Return the findings to the coordinator, citing each claim's source.
  3. The coordinator writes the returned findings to a single Markdown file. Save it where the repo already keeps such notes; match the existing convention, and if there is none, put it somewhere sensible and say where.
Files (skills)
  • agents
    • openai.yaml 246 B
      interface:
        display_name: "Research"
        short_description: "Research primary sources through a selected pack adapter"
        default_prompt: "Use $research to investigate this question against primary sources and save the findings in the repository."
      
  • SKILL.md 3.6 KB
    ---
    name: research
    description: Investigate a question against high-trust primary sources and capture the findings as a Markdown file in the repo. Use when the user wants a topic researched, docs or API facts gathered, or reading legwork delegated to a research worker.
    ---
    
    ## Research-worker dispatch
    
    Use exactly one research worker; never create a chain of workers.
    
    The coordinator supplies the selected adapter, explicit research-worker model,
    explicit research-worker effort, and the complete research task. Dispatch only
    through `skills/drivers/orchestrate/scripts/codex-worker.py` or
    `skills/drivers/orchestrate/scripts/claude-worker.py`, using the selected
    adapter's read-only surface. Never use the built-in Agent tool, Workflow tool,
    or background-agent machinery.
    
    Use the provider-neutral hosted-source capability defined by the selected
    adapter's dispatch reference on every read-only research `start`, with one
    coordinator-owned state file under the coordinator's session-scratch directory.
    The selected adapter's `resume`, `stop`, and `verify` surfaces retain lifecycle
    ownership. Do not restate provider argv here.
    
    After selection, this interface does not reclassify research work or choose a
    model or effort. Preserve adapter-owned state and coordinator-owned recovery
    through the orchestrate adapter contract; do not restate its command or
    lifecycle mechanics here.
    
    Before dispatch, the coordinator writes the exact complete research-task prompt
    bytes to an immutable session-scratch file and passes that file's contents as
    the selected adapter's positional prompt. The worker receives no chat transcript
    or other coordinator-session material. The prompt requires the worker to return
    a final message beginning `SOURCE_ACCESS_UNAVAILABLE:` when hosted search or
    fetch is refused, and to write no findings document in that case.
    
    Only when that start completed with a terminal, resumable session and its final
    message begins with the exact `SOURCE_ACCESS_UNAVAILABLE:` prefix may the
    coordinator fetch the required public sources. Create a unique
    `.research-sources.*` directory under the worker cwd, store only fetched public
    unauthenticated source files plus `manifest.md` mapping each file to its source
    URL, and resume the same worker with the manifest's absolute path. This bounded
    handoff is the sole narrow exception to the original-prompt-only rule and takes
    precedence over it. The directory is removed after completion.
    
    If the adapter leaves no resumable session or the coordinator cannot obtain the
    required primary sources, stop explicitly and write no successful findings
    document. Never put transcripts, credentials, secrets, `.env`, patient data,
    authenticated or private source material, or unrelated coordinator-session
    content in the handoff.
    
    The worker performs the research directly and returns source-cited findings to
    the coordinator. Never spawn another background agent or nested worker. If the
    worker fails or is interrupted, report the failure explicitly. Do not describe a
    successful dispatch as completed research. The coordinator writes the returned
    findings to the single Markdown file required below.
    
    ## The research worker's job
    
    1. Investigate the question against **primary sources** — official docs, source code,
       specs, first-party APIs — not a secondary write-up of them. Follow every claim back to
       the source that owns it.
    2. Return the findings to the coordinator, citing each claim's source.
    3. The coordinator writes the returned findings to a single Markdown file. Save it where
       the repo already keeps such notes; match the existing convention, and if there is none,
       put it somewhere sensible and say where.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.

Related