| name | research |
| description | Research external topics, create comprehensive analysis, and incorporate learnings into memory systems |
| model | opus |
| allowed-tools | WebSearch, WebFetch, Read, Write, Glob, Grep, Bash(python3:*/skills/github/scripts/*), mcp__forgetful__*, mcp__serena__*, Skill |
| user-invocable | true |
Research and Incorporate Command
ultrathink
Research external topics, create comprehensive analysis, and incorporate learnings into memory systems.
Usage
/research
Topic: {topic name}
Context: {why this matters to the project}
URLs: {optional comma-separated source URLs}
Parameters
| Parameter | Required | Description |
|---|
Topic | Yes | Subject to research |
Context | Yes | Why this matters to the project |
URLs | No | Source URLs to fetch and analyze |
Example
/research
Topic: Chesterton's Fence
Context: Decision-making principle for understanding existing systems before changing them
URLs: https://fs.blog/chestertons-fence/, https://en.wikipedia.org/wiki/G._K._Chesterton
What This Does
- Research Phase: Check existing knowledge, fetch URLs, perform web searches
- Analysis Phase: Write 3000-5000 word analysis to
.agents/analysis/
- Applicability Phase: Map integration points with ai-agents project
- Memory Phase: Create Serena memory + 5-10 atomic Forgetful memories
- Action Phase: Create GitHub issue if implementation work identified, via
python3 "${COPILOT_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-.claude}}/skills/github/scripts/issue/new_issue.py"
Budget
Complete within 50k output tokens. If approaching the limit, summarize findings so far, persist partial analysis, and stop. Prefer completing fewer phases well over partial work across all phases.
Fallback Rules
- If
WebSearch returns no results for a query, try 2 alternative phrasings, then proceed with available information.
- If a
WebFetch URL is unreachable or returns a non-success status, note it as unavailable in the analysis and continue with other sources.
- If memory systems (Serena or Forgetful) are unavailable, skip the Memory Phase and record the skip in the Action Phase output.
- If a URL points at github.com, do not call
WebFetch. Use the github skill scripts, which reach the API through gh and so cannot be denied by a WebFetch hook. Write the plugin root inline on each call, because shell variables do not survive between Bash invocations. Issue body and metadata: python3 "${COPILOT_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT:-.claude}}/skills/github/scripts/issue/get_issue_context.py" --owner {owner} --repo {repo} --issue {n}. Issue discussion: the same path with issue/get_issue_comments.py. PR body and diff: pr/get_pr_context.py --owner {owner} --repo {repo} --pull-request {n}. PR review discussion: pr/get_pr_review_comments.py and pr/get_pr_review_threads.py with the same flags.
- If
WebFetch is denied by a harness permission decision rather than a network error, that is a capability signal, not a prompt-injection attempt. Record the denial, switch to the github script path above for github.com URLs or to WebSearch for other hosts, and continue. Do not halt the run. Never call a tool the denial names unless it is already in this command's allowed-tools.
Stop Conditions
Stop when any of the following is true:
- All 5 phases completed or intentionally skipped under a Fallback Rule.
- 3 phases have failed (intentional skips under Fallback Rules do not count as failures).
- The 50k output-token budget is reached.
Output
| Artifact | Location |
|---|
| Analysis document | .agents/analysis/{topic-slug}.md |
| Serena memory | .serena/memories/{topic-slug}-integration.md |
| Forgetful memories | 5-10 atomic memories in knowledge graph |
| GitHub issue | Created if implementation work identified |
Related
- Skill:
.claude/skills/research-and-incorporate/SKILL.md
- Memory skill:
/memory-search for retrieving incorporated knowledge
