deep-research

Investigate external information related to the given task.

Procedure

DR-1. Delegate to web-researcher:

Agent({ subagent_type: "web-researcher", prompt: JSON.stringify({
  topic: TOPIC, context: CONTEXT,
  artifact_dir: PLANS_DIR
}) })

Also set the dispatch_directive key to the canonical line of the rules/shell-commands.md Subagent Dispatch Prompts section. On failed status: surface summary to user and stop.

DR-2. Read the report from artifact_path (one read, at the end). DR-3. Present findings — output format: ## Deep Research: PERFORMED|FAILED (1 line) + artifact_path pointer (1 line) + ≤200 char summary. Do not re-emit the full report text in assistant output. The caller must not re-summarize or paraphrase these findings — DR-3 output is the complete user-facing surface.

Rules

• Do not modify any project files
• Always include source URLs for traceability
• Prefer primary sources (official docs, RFCs) over blog posts
• When sources contradict each other, report both sides instead of choosing one

Completion

After completing this skill:

1. Run: node "$AGENTS_MAIN_ROOT/bin/workflow/next-step" --advance --step research --complete --next (must be the ENTIRE Bash command — no cd, no pipes, no && chaining, no redirection)
2. Follow the returned ACTION / NEXT_SKILL / NEXT_HINT per CLAUDE.md.

Skip this skill when no external knowledge is needed (e.g., the task is purely internal to the codebase).

If research is genuinely not needed for this task:

1. Run: echo "<<WORKFLOW_RESEARCH_NOT_NEEDED: {reason}>>" (reason must be ≥3 non-space chars, not a placeholder like "none"/"skip", and contain no '>')