tq-forge-agents

name: tq-forge-agents description: | List every forged agent across the sandbox and production with kind, current quality score, hand_off_to wiring, and a one-line capability summary. Read-only. Use when asked for "/tq-forge-agents", "show agents", "list my agents", or "what does this agent do". allowed-tools: - Bash - Read

/tq-forge-agents — agent registry view

When to use

You want a snapshot of every forged agent — sandbox + promoted — with its kind (researcher / coder / etc.), current quality score, the agents it can hand off to, and a one-line capability summary from AGENT.md. This is the agent-only counterpart of /tq-forge-list, which mixes skills + agents.

Procedure

export TQ_FORGE_HOME="${TQ_FORGE_HOME:-$HOME/.tq-forge}"
source "${CLAUDE_PLUGIN_ROOT:-${TQ_FORGE_HOME:-$HOME/.tq-forge}/install}/scripts/common.sh" && tq_ensure_home

1. Enumerate agent directories.

   find "$SANDBOX_AGENTS" "$CLAUDE_SKILLS_DIR/agents" -maxdepth 2 -name AGENT.md 2>/dev/null
   

2. For each agent, read metadata — yaml + json in one pass:

   python3 - <<'PY'
   import json, re, pathlib, os
   roots = [pathlib.Path(os.environ['SANDBOX_AGENTS']),
            pathlib.Path(os.environ['CLAUDE_SKILLS_DIR'])/'agents']
   for r in roots:
       if not r.exists(): continue
       for d in sorted(p for p in r.iterdir() if p.is_dir()):
           agm, tj = d/'AGENT.md', d/'tools.json'
           if not agm.exists(): continue
           text = agm.read_text(errors='replace')
           kind = (re.search(r'kind:\s*(\S+)', text) or [None,'?'])[1]
           tools = hand = []
           if tj.exists():
               try:
                   j = json.loads(tj.read_text())
                   tools, hand = j.get('tools',[]), j.get('hand_off_to',[])
               except Exception: pass
           loc = 'sandbox' if str(d).startswith(os.environ['SANDBOX_AGENTS']) else 'promoted'
           names = [h if isinstance(h,str) else h.get('name','?') for h in hand]
           print(f"{d.name}\t{kind}\t{loc}\t{len(tools)}\t{','.join(names) or '-'}")
   PY
   

3. Pull current scores for each agent slug from skill-log.json:

   python3 -c "import json,pathlib,os;d=json.loads((pathlib.Path(os.environ['TQ_FORGE_HOME'])/'skill-log.json').read_text() or '[]');print(json.dumps({e['slug']:e.get('score') for e in d if e.get('kind')=='agent'}))"
   

4. Render the table.

   AGENT               KIND        LOC       SCORE  TOOLS  HAND-OFF
   market-research     researcher  promoted  🟢 8.7    3    business-analyst, reviewer
   inbox-triage        ops-manager sandbox   🟡 6.5    3    coder
   

   Color: 🟢 >= 8, 🟡 >= 6, 🔴 < 6. Show "—" if score missing.

5. Print a totals footer.

   <N> agents · <S> sandbox · <P> promoted · avg score <X.X>/10
   

6. Suggest follow-ups only when applicable: sandbox agents >=7 -> /tq-forge-promote; promoted agents <7 -> /tq-forge-improve; none found -> /tq-forge-agent <intent>.

Pitfalls

• An agent with missing or malformed AGENT.md yaml trips the parser. Skip it with a warning rather than aborting the listing.
• hand_off_to may be a list of strings or objects ([{"name":...,"when":...}]). Show the name field if it's an object.
• This view reads only forged agents under the two known roots — it won't show unrelated subagents the host may have.

Verification

• The table has one row per AGENT.md found under the two roots.
• Total count matches find ... -name AGENT.md | wc -l.

Tags

tq-forge agents registry read-only