ClawGear Developer Persona

Type: Personas
Version: 1.0.0
Price: $29

AI can write code fast. That's not the same as AI that ships code correctly.

This persona installs a developer ops layer: Claude Code dispatch protocol, PR gate checks before every merge, daily standup auto-generated from git log, and a hard staging-before-production rule. The agent does the grunt work — code review, standup, routine staging deploys. You decide what goes to production.

Setup time: under 30 minutes.

---

What You Get

SOUL.md          — Developer identity — shipping-focused, skeptical of AI output, staging gate rules
AGENTS.md        — Claude Code dispatch protocol, PR gate procedure, deploy rules
HEARTBEAT.md     — PR queue check, CI status, failed tests alert
cron-schedule.md — Standup, nightly build check, PR review queue crons
install.md       — Setup walkthrough including Claude Code and GitHub CLI configuration
README.md        — Overview and prerequisites

---

What It Does

• Claude Code dispatch — spawns Claude Code sub-agents for coding tasks from chat or cron; checks loop guard before dispatching to avoid re-trying approaches that already failed
• PR gate — runs before every merge: hardcoded secrets scan, file count check (<20 changed files), TODO/FIXME delta, test existence check, build pass verification
• Daily standup — reads git log and TASKS.md, writes standup at 9 AM in the format: "shipped yesterday / ships today / blocked on" — not a vague status update, a forcing function
• Deployment gate — agents push to staging; you promote to production via explicit human review; no AI ever deploys directly to production
• Loop guard — before any Claude Code dispatch, checks scripts/loop-guard.py for previously tried/blocked approaches; searches memory for what failed; dispatches with a different strategy
• CI monitoring — heartbeat checks for failed tests, broken builds, PR queue depth; Telegram alert if anything is red

---

Who It's For

Developers and small engineering teams who want AI in the workflow without AI making deployment decisions.

The agent handles everything up to staging. You hold the merge-to-main and production deploy keys. That's intentional.

Good fit if:

• You use Claude Code (or want to) and want a structured dispatch protocol
• PRs are getting merged too fast without review — you want a mechanical gate
• You hate writing daily standups from scratch
• You want CI/PR alerts without setting up a full monitoring stack

Not a fit if:

• You want fully autonomous deployments (this persona explicitly rejects that)
• You don't use GitHub (PR gate requires gh CLI)
• You don't have Claude Code installed (~/.npm-global/bin/claude)

---

Install

# Prerequisites
gh auth status          # GitHub CLI authenticated
claude --version        # Claude Code installed
openclaw gateway status # OpenClaw running

# 1. Copy persona files to workspace reference
cp -r /path/to/skill/files ~/.openclaw/workspace/developer-ref

# 2. Copy core files to workspace root
cp developer-ref/SOUL.md ~/.openclaw/workspace/SOUL.md
cp developer-ref/AGENTS.md ~/.openclaw/workspace/AGENTS.md
cp developer-ref/HEARTBEAT.md ~/.openclaw/workspace/HEARTBEAT.md

# 3. Replace placeholders in SOUL.md
#    {{AGENT_NAME}}     — your agent's name
#    {{OPERATOR_NAME}}  — your name
sed -i '' \
  -e 's/{{AGENT_NAME}}/YourAgentName/g' \
  -e 's/{{OPERATOR_NAME}}/YourName/g' \
  ~/.openclaw/workspace/SOUL.md

# 4. Create workspace files
touch ~/.openclaw/workspace/MEMORY.md
touch ~/.openclaw/workspace/TASKS.md
mkdir -p ~/.openclaw/workspace/scripts

# 5. Install loop guard script (see install.md for full source)
cat developer-ref/install.md

# 6. Wire up cron jobs
cat developer-ref/cron-schedule.md

# 7. Restart gateway
openclaw gateway restart

See install.md for the full walkthrough including GitHub token setup, loop guard script, and first standup verification.

---

PR Gate Checks

Every PR before merge runs this checklist:

| Check | Pass criteria | |---|---| | Hardcoded secrets | No matches for key, secret, password, token in diff | | File count | < 20 changed files (larger PRs get flagged to break up) | | TODO/FIXME count | Count hasn't increased from base branch | | Test coverage | New functions have corresponding test files | | Build | npm run build (or equivalent) passes |

If any check fails — PR is blocked, Telegram alert sent, reason logged.

---

Claude Code Dispatch Protocol

Task arrives -> check loop-guard.py (was this tried before?)
  -> IF blocked: search memory for failure reason, pick different approach
  -> IF clear: spawn Claude Code with:
      - context files (MEMORY.md, relevant code)
      - success criteria (explicit)
      - test requirements
      - output format expected
  -> Wait for completion
  -> Run PR gate on output
  -> IF passes: merge to staging branch
  -> Surface to operator for production review

Never merge to main directly. Never skip the gate.

---

Standup Format

Generated every morning at 9 AM from git log and TASKS.md:

Daily Standup — [date]

Shipped yesterday:
- [commit messages or PR titles that merged]

Ships today:
- [tasks marked in-progress in TASKS.md]

Blocked:
- [anything stuck > 2 days with reason]

If "shipped yesterday" is empty for 3 consecutive standups — flag sent to Telegram.

---

Changelog

• 1.0.0 (2026-03-09): Initial ClawMart release.