analysis-agent

/analysis-agent

Run a specialist sub-agent for the current analysis stage.

Usage:

• /analysis-agent — show top recommended agents for current stage
• /analysis-agent 1 — run the #1 recommended agent
• /analysis-agent 1 3 — run agents #1 and #3 in parallel
• /analysis-agent all — run all recommended agents
• /analysis-agent "통계" — fuzzy match by alias and run

---

Instructions

Step 1: Load context

1. Read .analysis/status.md → find active Full analysis (or ask which if multiple)
2. Read .analysis/agent-state.md (if exists) → load suppression history
3. Read .analysis/agents.yml (if exists, else fallback to core/config/agents.yml) → user config
4. Determine current stage from analysis folder (which stage files exist)
5. Read the current stage file (all sections)

Step 2: Parse argument

No argument → go to Step 3 (show recommendations)

Number(s) (e.g., 1, 1 3, all) → go to Step 5 (execute directly)

• "all" = run all agents in the last recommendation block

Free text / alias (e.g., "통계", "hypothesis", "리뷰") →

• Fuzzy-match against id, label, and aliases fields in core/agents/registry.yml
• If 1 match: run it directly → Step 5
• If 2+ matches: show compact list, ask user to pick (single question)
• If 0 matches: show "No match. Available agents for {stage}: {list}" and exit

Step 3: Evaluate router (no-arg flow)

Apply routing rules from core/agents/router.yml for current stage:

1. Check gates: evaluate gate auto_run conditions → list auto-run gates separately
2. Score optional agents: apply rule conditions + bonuses → filter by min_score_threshold
3. Apply suppression: skip agents in agent-state.md for this stage (unless content hash changed)
4. Apply user config: skip disabled agents, honor max_recos_per_stage
5. Take top_k: select top 3 by score

Step 4: Show recommendation block

─────────────────────────────────────────────────────
🤖 Specialist Recommendations — {STAGE} stage
  Analysis: {ID} — {title}
─────────────────────────────────────────────────────
{If auto-gates fired:}
  [auto] Running {gate-id} — {reason}

{Recommendations:}
  1. {label}  —  {reason_template from router.yml}
  2. {label}  —  {reason_template}
  3. {label}  —  {reason_template}
─────────────────────────────────────────────────────
Run? (1 / 2 / 3 / 1 3 / all / n)  →