tq-forge-test

name: tq-forge-test description: | Re-run dry-test + quality-score on an existing forged skill or agent without rewriting it. Useful after a manual edit, or to verify a flagged "needs_manual_review" item now passes. Use when asked for "/tq-forge-test", "re-test the skill", "score it again", or "is this still passing". allowed-tools: - Bash - Read

/tq-forge-test — re-validate without rewriting

When to use

You manually edited a sandbox or production artifact and want to confirm it still passes. Or forge-queue.json has a slug in needs_manual_review and you want to know if today's edit cleared the bar. This skill never rewrites the artifact — only the score.

Procedure

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

1. Resolve the target. Accept a slug or a full path. With a slug, auto-detect across sandbox + production:

   bash "$S/quality-score.sh" --auto-detect "<slug>"
   

   If missing everywhere, list the sandbox + production dirs and abort:

   ls "$SANDBOX_SKILLS" "$SANDBOX_AGENTS" "$CLAUDE_SKILLS_DIR" 2>/dev/null
   

2. Run the structural dry-test (resolve the path first via tq_resolve):

   P="$(tq_resolve "<slug>")"; bash "$S/dry-test.sh" --json "$P"
   

   Capture verdict (pass/warn/fail), checks_passed, and issues.

3. Run quality-score.

   bash "$S/quality-score.sh" --json "$P"
   

   Capture avg and per-dimension breakdown.

4. Print a combined report. Two-line header, then the per-dimension bars, then any dry-test issues as bullets:

   <emoji> <slug> (<kind>) — quality X/10 · dry-test Y/10 (<verdict>)
   

5. Update skill-log.json with the new score (mutate only the matching entry):

   python3 -c "import json,datetime,pathlib,os;p=pathlib.Path(os.environ['TQ_FORGE_HOME'])/'skill-log.json';d=json.loads(p.read_text() or '[]');[e.update({'score':<new>,'last_tested':datetime.datetime.now(datetime.timezone.utc).isoformat()}) for e in d if e.get('slug')=='<slug>'];p.write_text(json.dumps(d,indent=2))"
   

6. If the slug was in needs_manual_review and avg >= 7, remove it:

   python3 -c "import json,pathlib,os;p=pathlib.Path(os.environ['TQ_FORGE_HOME'])/'forge-queue.json';q=json.loads(p.read_text());q['needs_manual_review']=[x for x in q.get('needs_manual_review',[]) if x.get('slug')!='<slug>'];p.write_text(json.dumps(q,indent=2))"
   

   Print: "✅ Cleared from needs_manual_review."

7. Suggest the next action. avg >= 7 -> suggest /tq-forge-promote. avg < 7 -> point at the lowest dimension and suggest /tq-forge-improve <slug>.

Pitfalls

• Check both copies if a slug exists in sandbox and production — the production copy is what users actually invoke. Test both.
• A skill that scored 9 last week but 6 now usually means an edit deleted a section. Run git diff (if the dir is tracked) to spot what changed.
• quality-score.sh is heuristic. 7-9 is the normal pass band; a perfect 10 is rare. Don't chase 10 by padding word count.

Verification

• Both subcommands exit 0 on success and non-zero on parse error — surface the exit code.
• skill-log.json has an updated last_tested for the slug.
• If the artifact was on needs_manual_review and now passes, that list is shorter by one.

Tags

tq-forge test re-score read-only