agentleFS
ric03uecric03uec/clawrium/.claude/skills/clawctl-uat/SKILL.md

clawctl-uat

Skill67 starsChanged 2 months agoApache-2.0View on GitHub

Run the daily end-to-end UAT for clawrium on a real host — install fresh agents of every type, verify existing agents survive, file (and close) a dated issue with the report

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run the daily lifecycle sweep for clawrium on a real host.

    line 9
  2. Distinctive:

    Before an on-call handoff, when you want to know the state of every managed host in one place.

    line 24
  3. Very specific:

    Don't run this more than once per day per host — the ephemeral agent names are namespaced by date (uatYYYYMMDD-*) and running twice will …

    line 27
  4. Distinctive:

    Only one UAT session per host per day.

    line 51
  5. Very specific:

    Every run files one GitHub issue titled daily live UAT run <date>.

    line 109
  6. Distinctive:

    After ~30 runs this tells you which host has the highest false-fail rate, which agent-type install path is flakiest, and which days …

    line 150
  7. Very specific:

    If the artifact dir for today exists but no issue was filed, the previous run died mid-flight.

    line 175

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Daily Live UAT

Run the daily lifecycle sweep for clawrium on a real host. Every day this skill installs one fresh agent per type, walks it through create → configure → start → exec, exercises the marquee behaviour checks, verifies pre-existing agents on the same host are undisturbed, cleans up, and files a dated GitHub issue containing the report.

This is a verification skill, not a build skill. It never changes source code. Its only side effects on the tree are .itx/uat-daily/<date>/ artifacts.

When to run

  • Daily, from cron or /schedule, at a fixed local time (default 10:00 PT so the operator sees the report over their morning coffee).
  • Manually after a significant merge, when you want a real-host smoke check before cutting a release.
  • Before an on-call handoff, when you want to know the state of every managed host in one place.

Don't run this more than once per day per host — the ephemeral agent names are namespaced by date (uatYYYYMMDD-*) and running twice will collide on agent create.

Modes

  • `` (no args) — run today's UAT in UTC. Default host: wolf-i.
  • <YYYY-MM-DD> — backfill for that date (uses the date literally in artifact paths and issue titles; still runs against the live fleet, so backfilling yesterday tests today's fleet with yesterday's label).
  • --host <alias> — override the default host. Useful for a per-host UAT sweep (clawctl:uat --host mac-test).

Preflight

Refuse to start unless:

git -C /home/devashish/workspace/ric03uec/clawrium rev-parse --is-inside-work-tree
uv run --directory /home/devashish/workspace/ric03uec/clawrium clawctl --version
uv run --directory /home/devashish/workspace/ric03uec/clawrium clawctl host get | grep -q <HOST>
command -v gh && gh auth status

Only one UAT session per host per day. Check for an existing artifact dir:

DATE=$(date -u +%Y-%m-%d)
if [ -d /home/devashish/workspace/ric03uec/clawrium/.itx/uat-daily/$DATE ]; then
  echo "Already ran today for $DATE. Delete .itx/uat-daily/$DATE/ to rerun."
  exit 0
fi

The loop

#StepOwner
0Preflight; render templateorchestrator
1Baseline — snapshot fleet, prove existing agents respondorchestrator
2Install — one fresh agent per type on HOSTorchestrator
3Configure + Start — walk each fresh agent to readyorchestrator
4Marquee — narrow behaviour checks per typeorchestrator
5Regression — re-run baseline; existing agents unchangedorchestrator
6Cleanup — delete fresh agents, diff against baselineorchestrator
7Write reportorchestrator
8File + close GitHub issue "daily live UAT run <date>"orchestrator
9Append ledger entryorchestrator

Full step-by-step in template.md (sibling file). This SKILL.md is the policy; template.md is the runbook.

Render + execute

The template lives at .claude/skills/clawctl-uat/template.md (or .opencode/skills/clawctl-uat/template.md, byte-identical). At the start of each run, substitute the placeholders and copy the rendered plan into the per-date artifact dir so the exact plan used for that run is preserved alongside the report:

DATE=$(date -u +%Y-%m-%d)
DATE_COMPACT=$(date -u +%Y%m%d)
HOST=${HOST:-wolf-i}
PROVIDER=${PROVIDER:-clm-openrouter}
PREFIX="uat${DATE_COMPACT}"
DIR=/home/devashish/workspace/ric03uec/clawrium/.itx/uat-daily/$DATE
mkdir -p "$DIR"

sed -e "s/{{DATE}}/$DATE/g" \
    -e "s/{{DATE_COMPACT}}/$DATE_COMPACT/g" \
    -e "s/{{HOST}}/$HOST/g" \
    -e "s/{{PROVIDER}}/$PROVIDER/g" \
    -e "s/{{PREFIX}}/$PREFIX/g" \
  .claude/skills/clawctl-uat/template.md > "$DIR/plan.md"

Then execute the plan phase-by-phase. Do not batch phases — each phase writes its own log so a mid-run abort leaves useful forensics.

Filing the issue

Every run files one GitHub issue titled daily live UAT run <date>. This is the searchable historical record. The issue body is the full report.

Close semantics:

  • PASS — close immediately with a comment noting the verdict. The issue exists for search + audit; there is nothing to action.
  • FAIL — leave open, apply agent-blocked label so it lands in the operator's queue.
  • PARTIAL (some phases PASS, some FAIL, cleanup succeeded) — leave open, no agent-blocked — the operator triages severity.
gh issue create \
  --title "daily live UAT run $DATE" \
  --label "authored-by:local_qwen" \
  --body-file "$DIR/report.md"

ISSUE_NUM=$(gh issue list --search "daily live UAT run $DATE in:title" \
              --state open --json number --jq '.[0].number')

case "$VERDICT" in
  PASS)    gh issue close $ISSUE_NUM --comment "PASS. Report body has full log." ;;
  FAIL)    gh issue edit $ISSUE_NUM --add-label "agent-blocked" ;;
  PARTIAL) : ;;
esac

Never create a new label. The two used above (authored-by:local_qwen, agent-blocked) already exist in the repo per the clawctl-lmwork skill's label discipline. If either goes missing, comment on the issue and stop.

Ledger

Append one line to .itx/uat-daily-ledger.jsonl in the main checkout (not in a worktree — this file is cross-run, and lives on main):

{"date":"2026-07-31","host":"wolf-i","verdict":"PASS","issue":123,"phases":{"preflight":"PASS","baseline":"PASS","install":"PASS","configure_start":"PASS","marquee":"PASS","regression":"PASS","cleanup":"PASS"},"anomalies":0,"ts":"2026-07-31T05:55:00Z"}

After ~30 runs this tells you which host has the highest false-fail rate, which agent-type install path is flakiest, and which days correlate with merges that need better test coverage.

Anomalies vs blockers

Anomalies (stale error hints, hidden flags, verb-drift) go in the report's Anomalies section as one-liners with severity BLOCKER / WARN / NOTE. Only BLOCKER counts against the phase verdict. A run with three WARN anomalies but every phase green is still PASS.

Non-goals

  • Not a replacement for make test. Unit tests catch code-shape bugs; this catches host-shape bugs.
  • Not a security review. Use ATX (atx review request) for that on the PR being verified.
  • Not a place to test speculative behaviour. Only checks landed on main belong in the marquee phase.
  • Not allowed to modify source code. If a UAT reveals a bug, file a separate issue for the bug; this skill's issue records only what was observed, not what to fix.

Recovery from a stuck run

If the artifact dir for today exists but no issue was filed, the previous run died mid-flight. Check .itx/uat-daily/<date>/ for the last log written, and look for orphaned uat<date>-* agents in clawctl agent get. Delete those first, then either delete the artifact dir to start clean or manually finish Phase 6 + reporting.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.