research

name	research
description	Ad-hoc research skill — investigate any topic (no IssueBoss issue required) by dispatching research-topic-processor, presenting findings inline, and optionally saving a document to .insights/shared/research/.
argument-hint	["topic or query"]

Research

You are using the research skill to investigate a free-form topic and present findings inline. No IssueBoss issue is required — this skill works for any ad-hoc query.

Trigger Phrases

This skill applies when:

• User says "research [topic]", "look into [topic]", "investigate [topic]", or "find out about [topic]"
• User invokes /research with a query
• Another skill needs free-form research outside the context of a specific issue

Inputs (from caller)

• topic (required) — the research question or subject; use the argument provided, or ask: "What would you like me to research?"
• tags (optional) — hashtag strings to steer dimension selection, e.g. ["#codebase", "#web"]. Valid tags: #codebase, #implementation, #architecture, #patterns, #conventions, #examples, #web, #external, #library, #practices, #web-deep. Pass [] if none are provided — the processor infers dimensions from the topic description.

VCS Context Resolution

Before dispatching the processor, resolve the current VCS context from the project root.

1. Check whether .jj/ exists in the project root.

2. If jj repo: Run:

   jj log -r @ --no-graph -T 'change_id ++ " " ++ commit_id'
   

   Split on the first space: first token = change_id, second token = commit.

3. If git repo (no .jj/): Run git rev-parse HEAD → commit; set change_id = "".

4. If neither: Set change_id = "" and commit = "".

Research Flow

1. Resolve VCS context — follow the VCS Context Resolution section above.

2. Dispatch issueboss:research-topic-processor via the Agent tool with this exact prompt format:

   Research the following topic:
   topic_description: {topic}
   topic_tags: {tags as JSON array, or [] if none}
   change_id: {change_id}
   commit: {commit}
   project_root: {absolute path to project root}
   

   Note: do NOT include an issue_slug line — this is an ad-hoc invocation.

3. Wait for the processor to return its structured output (delimited by ---FRONTMATTER---, ---FINDINGS---, ---END---).

4. Parse the output:

   • Extract the YAML block between ---FRONTMATTER--- and ---FINDINGS--- as frontmatter fields.
   • Extract everything between ---FINDINGS--- and ---END--- as the findings body.
   • If either delimiter is missing, inform the user: "The research processor returned unexpected output — findings cannot be displayed." Do not attempt to save or present any document.

5. Present findings inline — display the full findings body to the user as-is.

6. Offer to save — after presenting, ask:

   Want me to save this as a research document in .insights/shared/research/?

   Do not save automatically. Wait for an explicit yes. If the user declines or gives no clear answer, do not write any file.

Saving the Document (if user says yes)

1. Derive the filename from the frontmatter topic field:

   • Split topic into space-delimited tokens; take the first 6 (or all tokens if fewer than 6)
   • Lowercase everything
   • Replace any non-alphanumeric characters (including spaces) with hyphens
   • Collapse consecutive hyphens to a single hyphen
   • Strip leading and trailing hyphens
   • Result: {kebab-summary}

2. Write the file to .insights/shared/research/{kebab-summary}.md using the Write tool. Do NOT run mkdir — the Write tool creates parent directories automatically.

   File content format:

   ---
   topic: {topic value from frontmatter}
   date: {date value from frontmatter}
   status: {status value from frontmatter}
   change_id: {change_id value from frontmatter}
   commit: {commit value from frontmatter}
   dimensions_active: {dimensions_active value from frontmatter}
   ---
   {findings body exactly as returned by the processor}
   

3. Confirm the path to the user:

   Saved to .insights/shared/research/{kebab-summary}.md

Constraints

• No issueboss MCP dependency — do not call any mcp__issueboss__* tools.
• No .claude/issueboss.json — do not read project config; derive all context from the query and VCS.
• Never save automatically — always ask before writing any document.
• No issue_slug — never include an issue_slug line in the processor prompt for this skill.