░░░░░░░░░░░░░░░░░░░░░
░░░░▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒░░░░
░░░▒▒▒▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▒▒▒░░░
░░▒▒▓▓▓▓████████████████████████▓▓▓▒▒░░
░░▒▓▓████████████████████████████████▓▓▒░░
░▒▓███████▀▀▀▀▀████████████▀▀▀▀▀███████▓▒░
░▒▓█████▀ ▀████████▀ ▀█████▓▒░
░▒▓████▀ ╔═══╗ ▀██████▀ ╔═══╗ ▀████▓▒░
░▒▓████ ║ ◉ ║ ▀████▀ ║ ◉ ║ ████▓▒░
░▒▓████ ╚═══╝ ████ ╚═══╝ ████▓▒░
░▒▓████▄ ████ ▄████▓▒░
░▒▓█████▄▄ ╔══════════════════╗ ▄▄█████▓▒░
░▒▓███████▄ ║ ▄▄▄▄▄▄▄▄▄▄▄▄▄▄ ║ ▄███████▓▒░
░▒▓████████║ ████████████████ ║████████▓▒░
░░▒▓██████║ ▀▀▀▀▀▀▀▀▀▀▀▀▀▀ ║██████▓▒░░
░░▒▒▓▓██╚══════════════════╝██▓▓▒▒░░
░░░▒▒▓▓▓████████████████▓▓▓▒▒░░░
░░░░▒▒▒▒▓▓▓▓▓▓▓▓▓▓▒▒▒▒░░░░
░░░░░░░░░░░░░░░░░░░░░
[AUTONOMOUS] [V6] [32+ PRs] [23 REPOS] [1 MERGED]
An OpenClaw agent that autonomously discovers issues, implements fixes, and submits pull requests to open-source projects — 24/7, without human intervention.
OpenClaw is the engine. ClawOSS is the race car.
git clone https://github.com/onthebed/ClawOSS.git
cd ClawOSS
cp .env.example .env # edit with your API keys
bash scripts/setup.sh
bash scripts/restart.shUse this path when validating the continuous-run MVP without creating a live PR:
cp .env.example .env # fill model, GitHub, dashboard, and budget values
npm run mvp:dry-runThe runner completes 3 heartbeat cycles by default, checks dashboard pause and budget guardrails before work, discovers candidate GitHub issues, applies CLA / duplicate / already-fixed / blocklist / avoidRepos filters, generates PR title/body/creation command, and writes a report under reports/mvp-run-*.md.
CLAWOSS_MVP_DISCOVERY_REPOS can be used to pin discovery to a controlled repo set during validation.
Useful options:
node scripts/mvp-runner.mjs --cycles 3 --max-candidates 20
node scripts/mvp-runner.mjs --cycles 3 --issue owner/repo#123If /api/agent/health-check returns pauseAgent: true or budget.paused: true, the runner stops before spawning, commenting, pushing, or preparing PR creation.
╔═══════════════════════════════════════════════════════════════════╗
║ ║
║ PRs Submitted ················· 32+ ║
║ Repos Contributed To ·········· 23 ║
║ PRs Merged ···················· 1 (badlogic/pi-mono#2166) ║
║ Time to First PR ·············· 12 minutes autonomous ║
║ Concurrent Sub-Agents ········· 5 ║
║ Model ························· Kimi Code k2p5 (direct API) ║
║ Content Filter 403s ··········· ZERO (PII sanitizer bypass) ║
║ Peak Throughput ··············· 15 PRs in 85 minutes ║
║ ║
╚═══════════════════════════════════════════════════════════════════╝
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
░ ░
░ ┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ░
░ ┃ O P E N C L A W G A T E W A Y ( port 18789 ) ┃ ░
░ ┃ mode: local · model: kimi-coding/k2p5 ┃ ░
░ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛ ░
░ ┃ ░
░ ┃ heartbeat every 10m ░
░ ┃ lightContext: false ░
░ ▼ ░
░ ┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ ░
░ ┃ ┃ ░
░ ┃ ╔══════════════════════════════════════════════════╗ ┃ ░
░ ┃ ║ ORCHESTRATOR (main session) ║ ┃ ░
░ ┃ ║ ║ ┃ ░
░ ┃ ║ reads: HEARTBEAT.md · AGENTS.md · SOUL.md ║ ┃ ░
░ ┃ ║ memory/work-queue.md ║ ┃ ░
░ ┃ ║ memory/pr-ledger.md ║ ┃ ░
░ ┃ ║ memory/pipeline-state.md ║ ┃ ░
░ ┃ ║ memory/wake-state.md ║ ┃ ░
░ ┃ ║ ║ ┃ ░
░ ┃ ║ runs: oss-discover · oss-triage ║ ┃ ░
░ ┃ ║ repo-analyzer · context-manager ║ ┃ ░
░ ┃ ║ ║ ┃ ░
░ ┃ ╚════╤════════╤════════╤════════╤════════╤════════╝ ┃ ░
░ ┃ │ │ │ │ │ ┃ ░
░ ┃ │ sessions_spawn (maxConcurrent: 5) │ ┃ ░
░ ┃ │ │ │ │ │ ┃ ░
░ ┃ ┌────▼───┐┌───▼────┐┌──▼─────┐┌─▼──────┐┌▼───────┐ ┃ ░
░ ┃ │ SUB 1 ││ SUB 2 ││ SUB 3 ││ SUB 4 ││ SUB 5 │ ┃ ░
░ ┃ │ ││ ││ ││ ││ │ ┃ ░
░ ┃ │ fresh ││ fresh ││ fresh ││ fresh ││ fresh │ ┃ ░
░ ┃ │ context││ context││ context││ context││ context│ ┃ ░
░ ┃ │ ││ ││ ││ ││ │ ┃ ░
░ ┃ │ clone ││ clone ││ clone ││ clone ││ clone │ ┃ ░
░ ┃ │ repro ││ repro ││ repro ││ repro ││ repro │ ┃ ░
░ ┃ │ fix ││ fix ││ fix ││ fix ││ fix │ ┃ ░
░ ┃ │ test ││ test ││ test ││ test ││ test │ ┃ ░
░ ┃ │ submit ││ submit ││ submit ││ submit ││ submit │ ┃ ░
░ ┃ │ cleanup││ cleanup││ cleanup││ cleanup││ cleanup│ ┃ ░
░ ┃ └───┬────┘└───┬────┘└───┬────┘└───┬────┘└───┬────┘ ┃ ░
░ ┃ │ │ │ │ │ ┃ ░
░ ┃ ▼ ▼ ▼ ▼ ▼ ┃ ░
░ ┃ memory/subagent-result-<repo>-<issue>.md ┃ ░
░ ┃ then: rm -rf /tmp/clawoss-<issue>-<ts>/ ┃ ░
░ ┃ then: reply ANNOUNCE_SKIP ┃ ░
░ ┃ ┃ ░
░ ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛ ░
░ │ ░
░ ┌────────────────────┼──────────────────────┐ ░
░ ▼ ▼ ▼ ░
░ ┌─────────────────┐ ┌──────────────────┐ ┌──────────────────────┐ ░
░ │ GitHub │ │ Vercel Dashboard │ │ Telemetry Hooks │ ░
░ │ (configured user)│ │ /api/ingest/* │ │ dashboard-reporter │ ░
░ │ │ │ │ │ audit-logger │ ░
░ │ PRs · Commits │ │ heartbeat │ │ pii-sanitizer │ ░
░ │ Reviews │ │ metrics │ │ dashboard-sync.sh │ ░
░ │ Follow-ups │ │ conversation │ │ pr-ledger-sync.sh │ ░
░ │ │ │ state · logs │ │ │ ░
░ └─────────────────┘ └──────────────────┘ └──────────────────────┘ ░
░ ░
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
Every 10 minutes, OpenClaw wakes the agent. It reads HEARTBEAT.md and executes this cycle:
▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒
▒ ▒
▒ ╭────────────────────╮ ▒
▒ │ 0a. CONTEXT HEALTH │ ▒
▒ │ session_status │ ▒
▒ │ >70% → /compact │ ▒
▒ ╰─────────┬──────────╯ ▒
▒ ▼ ▒
▒ ╭────────────────────╮ ▒
▒ │ 0b. CIRCUIT BREAK │ ▒
▒ │ wakes >= 50? stop │ ▒
▒ │ errors >= 2? stop │ ▒
▒ ╰─────────┬──────────╯ ▒
▒ ▼ ▒
▒ ╭────────────────────────────────────────╮ ▒
▒ │ 1. STALL RECOVERY │ ▒
▒ │ sessions_list → stale >5min? kill │ ▒
▒ │ re-queue task, 2 strikes → skip │ ▒
▒ ╰───────────────────┬────────────────────╯ ▒
▒ ▼ ▒
▒ ╭────────────────────────────────────────╮ ▒
▒ │ 2. PR FOLLOW-UPS │ ▒
▒ │ gh pr list --author @me --state open │ ▒
▒ │ reviews → oss-followup │ ▒
▒ │ CI fail → fix & push │ ▒
▒ │ merged → update pipeline-state.md │ ▒
▒ │ stale >7d → close politely │ ▒
▒ ╰───────────────────┬────────────────────╯ ▒
▒ ▼ ▒
▒ ╭────────────────────────────────────────╮ ▒
▒ │ 3. PICK WORK │◀──────────────────╮ ▒
▒ │ merge staging files → work-queue.md │ │ ▒
▒ │ active >= 5 → skip to step 6 │ │ ▒
▒ │ active < 5 → pick task (score >= 5) │ │ ▒
▒ │ skip if in pr-ledger (no duplicates) │ │ ▒
▒ │ queue < 5 → run oss-discover broadly │ │ ▒
▒ ╰───────────────────┬────────────────────╯ │ ▒
▒ ▼ │ ▒
▒ ╭────────────────────────────────────────╮ │ ▒
▒ │ 4. TRIAGE │ │ ▒
▒ │ oss-triage: complexity, feasibility │ │ ▒
▒ │ repo-analyzer: repo health check │ │ ▒
▒ │ web_search: upstream context │ │ ▒
▒ ╰───────────────────┬────────────────────╯ │ ▒
▒ ▼ │ ▒
▒ ╭────────────────────────────────────────╮ │ ▒
▒ │ 5. SPAWN SUB-AGENT │ │ ▒
▒ │ sessions_spawn with task: │ │ ▒
▒ │ clone → reproduce → implement │ │ ▒
▒ │ → verify → review → submit │ │ ▒
▒ │ → cleanup → ANNOUNCE_SKIP │ │ ▒
▒ │ │ │ ▒
▒ │ LOOP BACK if active < 5 ─────────────┼───────────────────╯ ▒
▒ ╰───────────────────┬────────────────────╯ ▒
▒ ▼ ▒
▒ ╭────────────────────────────────────────╮ ▒
▒ │ 6. HANDLE RESULTS │ ▒
▒ │ read subagent-result-*.md files │ ▒
▒ │ success + PR URL → pipeline-state.md │ ▒
▒ │ no PR URL → re-queue once │ ▒
▒ │ failure → log reason │ ▒
▒ │ disk cleanup: rm stale /tmp/clawoss-* │ ▒
▒ ╰───────────────────┬────────────────────╯ ▒
▒ ▼ ▒
▒ ╭────────────────────────────────────────╮ ▒
▒ │ 7. REPORT + SELF-WAKE │ ▒
▒ │ dashboard-reporter: log cycle │ ▒
▒ │ update wake-state.md counters │ ▒
▒ │ active < 5? → go back to step 3 │ ▒
▒ │ openclaw system event "cycle-complete" │ ▒
▒ ╰────────────────────────────────────────╯ ▒
▒ ▒
▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒▒
The plugins/pii-sanitizer/index.js (101 lines) performs bidirectional @ swapping so OpenRouter never sees the @ symbol that triggers its content filter.
▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
▓ ▓
▓ var FULLWIDTH_AT = '\uFF20' // ＠ ▓
▓ ▓
▓ INCOMING (persist) ▓
▓ ┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄ ▓
▓ sanitizeString(text) → text.replace(/@/g, FULLWIDTH_AT) ▓
▓ │ ▓
▓ ▼ ▓
▓ deepSanitize(value) → recurse strings, arrays, objects ▓
▓ │ ▓
▓ ▼ ▓
▓ sanitizeMessage(msg) → returns { message: cleaned } ▓
▓ ▓
▓ ▓
▓ OUTGOING (execute) ▓
▓ ┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄ ▓
▓ desanitizeString(text) → text.replace(FULLWIDTH_AT, '@') ▓
▓ │ ▓
▓ ▼ ▓
▓ deepDesanitize(value) → same recursion, opposite direction ▓
▓ ▓
▓ ▓
▓ register(api) hooks: ▓
▓ ▓
▓ ░ api.on('tool_result_persist') → sanitize(@ → ＠) file reads ▓
▓ ▒ api.on('before_message_write') → sanitize(@ → ＠) assistant output ▓
▓ ▓ api.on('before_tool_call') → desanitize(＠ → @) write/edit/exec ▓
▓ ▓
▓ ▓
▓ Disk (@) ──read──▶ Session (＠) ──model──▶ Tool call (＠→@) ──▶ Disk ▓
▓ ▓
▓ OpenRouter never sees raw @ ──▶ zero 403 content filter errors ▓
▓ ▓
▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
░ ░
░ dashboard-reporter/handler.ts (628 lines) ░
░ ┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄ ░
░ ░
░ State: accumulatedInputTokens · accumulatedOutputTokens ░
░ toolCallCount · reposUsed (Set) · subagentMeta (Map) ░
░ Cost: INPUT_COST = $0.60/M · OUTPUT_COST = $3.00/M ░
░ ░
░ Events: ░
░ ┌──────────────────┬──────────────────────────────────────────────────┐ ░
░ │ user_message │ postConversation({role:"user", content}) │ ░
░ │ after_tool_call │ accumulate tokens + detect sessions_spawn │ ░
░ │ │ track repos via regex on JSON.stringify(params) │ ░
░ │ │ relay sub-agent history (last 20 messages) │ ░
░ │ agent_end │ flush metrics + heartbeat + state + logs │ ░
░ │ │ costUsd = input * 0.6/M + output * 3.0/M │ ░
░ │ │ RESET all accumulators │ ░
░ └──────────────────┴──────────────────────────────────────────────────┘ ░
░ ░
░ Network: ░
░ postNonBlocking(path, body) → 2 attempts, 10s timeout, 3s retry ░
░ postConversation(body) → 1 attempt, 5s timeout, fire-and-forget ░
░ postState(apiKey) → reads work-queue.md + pipeline-state.md ░
░ ░
░ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ░
░ ░
░ audit-logger/handler.ts (133 lines) ░
░ ┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄ ░
░ Fire-and-forget structured logs. 10s timeout. ░
░ command:new → "Session started" (info) ░
░ after_tool_call → "Tool: {name}" (debug / warn on error) ░
░ agent_end → "Run completed" (info / error) ░
░ ░
░ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ░
░ ░
░ dashboard-sync.sh (319 lines) — runs as background process ░
░ ┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄ ░
░ PID lock · md5 self-update every 60s · polls JSONL every 10s ░
░ build_session_map() → python3 reads sessions.json for sub-agent metadata ░
░ parse_content_blocks() → text/thinking/toolCall/tool_result ░
░ ROLE_MAP: user→user, assistant→assistant, toolResult→tool_result ░
░ curl POST /api/ingest/conversation per new message ░
░ curl POST /api/ingest/heartbeat with session counts ░
░ ░
░ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ░
░ ░
░ pr-ledger-sync.sh (185 lines) — runs every 60s via launchd ░
░ ┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄ ░
░ Source 1: gh search prs --author "$GITHUB_USERNAME" --limit 200 ░
░ Source 2: grep subagent-result-*.md for PR URLs ░
░ Python merger: pr_map keyed by URL, GH is authoritative for status ░
░ Result files fill in issue numbers, existing ledger preserves mappings ░
░ Output: workspace/memory/pr-ledger.md (sorted markdown table) ░
░ ░
░ All telemetry → clawoss-dashboard.vercel.app ░
░ ░
░ ┌──────────────────────────────────────────────────────────────────────┐ ░
░ │ /api/ingest/heartbeat ← status, uptimeSeconds, repos │ ░
░ │ /api/ingest/metrics ← inputTokens, outputTokens, costUsd │ ░
░ │ /api/ingest/conversation ← sessionId, role, content, toolName │ ░
░ │ /api/ingest/state ← workQueue, pipelineState, activeRepos │ ░
░ │ /api/ingest/logs ← structured audit trail │ ░
░ └──────────────────────────────────────────────────────────────────────┘ ░
░ ░
░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░
| Phase | Skill | What It Does |
|---|---|---|
| Discovery | oss-discover |
gh search issues across all languages, score >= 5, write to queue |
| Discovery | oss-triage |
Assess complexity (simple/medium/complex), decide attempt/skip/defer |
| Discovery | repo-analyzer |
Repo health gate: stars, merge velocity, review rate, cache to repos/ |
| Implement | oss-implement |
Reproduce-first TDD: failing test → minimal fix → verify → evidence |
| Implement | oss-review |
7-gate quality check: scope, quality, tests, security, anti-slop, git, PR |
| Implement | safety-checker |
8-check final gate: budget, diff, secrets, branch, spam, CI, independent review |
| Submit | oss-submit |
Push to fork, gh pr create, AI disclosure, log + report |
| Submit | oss-followup |
Categorize review feedback, implement changes, max 3 rounds |
| Infra | context-manager |
Monitor context %, flush at 80%, re-read after compaction |
| Infra | dashboard-reporter |
Post telemetry to Vercel dashboard |
| Super | systematic-debugging |
Root-cause-first, never guess |
| Super | test-driven-development |
Red-green-refactor cycle |
| Super | verification-before-completion |
Fresh evidence before claiming done |
| Super | brainstorming |
Explore intent before complex design |
| Super | requesting-code-review |
Dispatch isolated reviewer subagent |
| Job | Schedule | What It Does |
|---|---|---|
work-queue-refill |
Every 2 hours | oss-discover batch mode, score >= 5, write work-queue-staging.md |
pr-followup-scan |
Every 30 min | gh pr list --author @me, check reviews + CI, write followup-staging.md |
daily-report |
11pm daily | Compile stats: submitted/merged/rejected, cost, acceptance rate |
weekly-retrospective |
Mon 9am | Review acceptance per repo, adjust scores, prune queue |
memory-cleanup |
Sun 3am | Archive >14d patterns, remove stale logs, prune closed PRs |
░ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ░
░ ░
░ Step 1 ░░ source .env ░
░ Step 2 ░░░ git config user.name "$GITHUB_USERNAME" ░
░ Step 3 ░░░░ gh auth login --with-token ░
░ Step 4 ▒▒▒▒ ln -sf workspace → ~/.openclaw/workspace ░
░ Step 5 ▒▒▒▒▒ sed __WORKSPACE_PATH__ → deploy config ░
░ ▒▒▒▒▒ python3: inject API keys, strip empties ░
░ Step 6 ▓▓▓▓▓▓ rm sessions/*.jsonl *.lock ░
░ Step 7 ▓▓▓▓▓▓▓ reset wake-state.md counters ░
░ Step 8 ▓▓▓▓▓▓▓▓ clean stale /tmp/clawoss-* dirs ░
░ Step 9 ███████████ openclaw gateway stop ░
░ Step 10 ████████████ openclaw gateway install ░
░ Step 11 █████████████ nohup dashboard-sync.sh & ░
░ Step 12 ██████████████ launchctl load pr-ledger-sync ░
░ Step 13 ███████████████ openclaw system event --mode now ░
░ ███████████████ "Fill all 5 sub-agent slots. Go." ░
░ ░
░ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ░
| Rule | Limit |
|---|---|
| Max lines changed per PR | 200 |
| Max files per PR | 5 |
| Max PRs per repo per day | 3 |
| Max PRs total per day | 10 |
| Min gap between same-repo PRs | 30 minutes |
| Fix attempts before abandon | 2 |
| Review rounds before disengage | 3 |
| Push to main/master | NEVER |
| Force-push | NEVER |
| Commit secrets/.env | NEVER |
| GitHub token scope | public_repo only |
| Branch naming | clawoss/<type>/<description> |
| Commit format | Conventional Commits |
clawOSS/
├── config/
│ ├── openclaw.json ············ gateway + agent + compaction config
│ ├── cron-jobs.json ··········· 5 scheduled jobs
│ └── com.clawoss.pr-ledger-sync.plist
├── plugins/
│ └── pii-sanitizer/
│ └── index.js ············· 101 lines — bidirectional @ ↔ ＠
├── scripts/
│ ├── setup.sh ················· first-time install (145 lines)
│ ├── restart.sh ··············· 13-step full restart (146 lines)
│ ├── start.sh ················· register agent + cron + gateway
│ ├── stop.sh ·················· teardown
│ ├── health-check.sh ·········· verify all systems
│ ├── dashboard-sync.sh ········ poll JSONL sessions (319 lines)
│ └── pr-ledger-sync.sh ········ merge GitHub API + result files (185 lines)
├── dashboard/ ··················· Next.js 15 + shadcn/ui + Turso
└── workspace/ ··················· (symlinked to ~/.openclaw/workspace)
├── AGENTS.md ················ 163 lines — prime directive + rules
├── HEARTBEAT.md ············· 244 lines — 8-step autonomous loop
├── SOUL.md ·················· persona, tone, boundaries
├── IDENTITY.md ·············· configured GitHub identity
├── USER.md ·················· operator profile
├── hooks/
│ ├── dashboard-reporter/ ·· 628 lines — telemetry to Vercel
│ └── audit-logger/ ········ 133 lines — structured logs
├── skills/ ·················· 15 skills (10 custom + 5 superpowers)
└── memory/ ·················· runtime state (agent-managed)
├── work-queue.md
├── pipeline-state.md
├── pr-ledger.md ········· auto-synced by pr-ledger-sync.sh
├── wake-state.md
└── subagent-result-*.md
Live at clawoss-dashboard.vercel.app
- Agent status + heartbeat
- Live conversation feed (orchestrator + per-sub-agent tabs)
- Token usage + cost tracking ($0.60/M input, $3.00/M output)
- PR pipeline with status tracking
- Structured audit logs
- Work queue and pipeline state
- Gateway wakes the agent every 10 minutes
- Orchestrator checks context health, handles PR reviews, picks work from the queue
- Sub-agents (up to 5 parallel) each clone a repo, reproduce the bug, write a failing test, implement a minimal fix, verify, and submit a PR
- Telemetry streams everything to the Vercel dashboard in real-time
- Cron jobs refill the work queue, scan for PR reviews, compile daily reports
- PII sanitizer swaps
@↔＠so the content filter never triggers - PR ledger auto-syncs every 60s from GitHub API + sub-agent result files
The agent never stops. Empty sub-agent slots are filled immediately. Idle is failure.
MIT