evaluate-team

name	evaluate-team
description	Team lead evaluates the current team's efficiency and composition: who is busy, who idle, whether a role is missing, whether someone can be let go. Produces a structured evaluation report to help the lead decide whether to /add-teammate or /remove-teammate. The agent can invoke autonomously (after natural-language user consent).
disable-model-invocation	false
allowed-tools	Read Glob Grep Bash

/evaluate-team — Team Efficiency and Composition Evaluation

Identity precheck

First infer from conversation context: if you're already clear you're a team lead (directory *_team/, with roundtable/), continue.

Only when unable to infer do on-disk check:

1. Glob _agent_team_work_zone/*_team/README.md

2. Locate the team workstation corresponding to the current conversation

3. If not a team lead → stop immediately and warn:

   ⚠️ /evaluate-team is only valid in team lead context.
   Currently not a team lead. If you need to assemble a team first, run /promote-to-team.
   

Phase 1: Collect current team state

1a. Read TEAMMATE_INFO.json (authoritative source)

Read _agent_team_work_zone/<SELF>_team/TEAMMATE_INFO.json — this is the authoritative source for current team roster; team_recipes/ is only historical audit.

Extract:

• active_teammates: each entry's name / role_source / model / scope / plan_mode_gating / spawned_at / last_checkpoint_at / revived_count / status
• offboarded_teammates: historical offboard records (reference only)

Pay special attention to:

• last_checkpoint_at > 24h ago → flag "checkpoint stale" (may be dysfunctional or abnormally idle)
• status == "failed_to_reactivate" → last /reactivate-team spawn failed; user must decide manual fix vs /remove-teammate
• status == "benched" → temporarily offline (full record + workstation retained); NOT counted as needing reactivate; list as "benched, wake on demand" with bench_reason (note: a stale last_checkpoint_at is expected for benched, not an anomaly)
• revived_count > 3 → repeated revival indicates instability; may be bad working-context.md or task unsuited for long running

If TEAMMATE_INFO.json doesn't exist or active_teammates is empty → team currently has no members; output "empty team" report and recommend /spawn-team.

Glob _agent_team_work_zone/<SELF>_team/team_recipes/*.md as historical supplement: read the latest recipe to understand what task the team was assembled for and the original design intent.

1b. Read roundtable/

Glob _agent_team_work_zone/<SELF>_team/roundtable/*.md, read frontmatter and content summary.

Count:

• Each teammate's recent submission count (aggregate by from field)
• Each teammate's OPEN/IN_PROGRESS task count
• Each teammate's latest activity time (latest file's date)

1c. Read ACTIVE_JOBS.md

Read _agent_team_work_zone/<SELF>_team/ACTIVE_JOBS.md to understand running long-term tasks (incl. tracker cron triggers).

Phase 2: Efficiency analysis

For each teammate, judge:

Dimension	Signal	Verdict
Busy	Frequent TASK / DONE / ERR submissions in roundtable (last 24h or per cron cadence); TODO has many items	Busy
Idle	No recent new output + no OPEN tasks on hand	Idle
Stuck	Has IN_PROGRESS with no follow-up update, or unresolved ERR	Stuck
Redundant	Two teammates have overlapping responsibilities (scopes in spawn prompt same or near-identical)	Redundant
Gap	Some type of work in the task decomposition has no one doing it (e.g. "needs smoke test" but no test-responsible teammate)	Missing role

Phase 3: Produce evaluation report

# Team Evaluation — <SELF> — YYYY-MM-DD HH:MM

## Current team composition
- Total: N
- Most recent /spawn-team: <timestamp>, recipe: <slug>

## Member status
| Nickname | Role source | Model | Status | Last Checkpoint | Revive | Recent output | Open on hand | Notes |
|---|---|---|---|---|---|---|---|---|
| Fixer | code-implementer | sonnet | Busy | 10 min ago | 0 | 4 DONE (last 24h) | 2 | Main line |
| Tracker | resources/agents/tracker.md | haiku | Idle | 2h ago | 0 | 0 | 0 | Awaiting next cron |
| Reviewer | resources/agents/reviewer.md | sonnet | Stuck | ⚠️ 36h ago | 1 | 1 IN_PROGRESS | 1 | Checkpoint stale + no work update |
| DevilAdvocate | resources/agents/devil-advocate.md | sonnet | Idle | 1h ago | 0 | 2 challenge reports | 0 | Initial critique done |

## Issues found

### 🚨 Immediate action
- Reviewer stuck 36h without update → suggest lead check its session for blocker
- No teammate responsible for final xlsx result output → missing result-reporter role

### ⚠️ Consider
- DevilAdvocate finished initial critique; no recent output → can /remove-teammate to free resources
  (unless decision-time check is needed later)

### ✅ Running well
- Fixer has stable pace
- Tracker outputs on cycle

## Suggested actions

### Immediate
1. Check Reviewer session for blocker (may be technical block or context pollution)
2. /add-teammate a result-reporter (start from resources/role_archetypes/analysis/result-reporter.md)

### Optional
- /remove-teammate DevilAdvocate (if initial critique fully absorbed)

## Token cost estimate
- Active teammates last 24h consumed approx X tokens (estimated from roundtable file size)
- If current pace continues for 1 week → approx X * 7 tokens

Phase 4: Request user confirmation

Do not auto-execute any action. Show report to user and ask:

What do you think of this evaluation? Which suggestion should I execute?
(Options: check Reviewer's blocker / add result-reporter / remove DevilAdvocate / other)

Based on user decision, call /add-teammate, /remove-teammate, or continue conversation.

Notes

• Evaluate only, don't execute: this skill's role is diagnosis and recommendation, not directly changing the team
• Imprecise quantification: token estimate is only a rough sense; no strict statistics
• Avoid over-optimization: don't manufacture work just to "keep each teammate loaded"; the team lead's role is to let tasks run smoothly, not to keep teammates busy
• Time window: defaults to last 24h activity; adjust based on task duration