Borrowing it
Nothing to install: this file belongs to saadnvd1/agent-os. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/saadnvd1/agent-os/main/.claude/agents/review-durability.mdgit clone --depth 1 https://github.com/saadnvd1/agent-osWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/saadnvd1/agent-os/review-durability)<a href="https://agentmods.dev/agents/saadnvd1/agent-os/review-durability"><img src="https://agentmods.dev/badge/agents/saadnvd1/agent-os/review-durability/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/saadnvd1/agent-os/review-durability"><img src="https://agentmods.dev/badge/agents/saadnvd1/agent-os/review-durability.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00070 | $0.01570 |
| Opus 5.5 | $0.00028 | $0.00628 |
| Sonnet 5.5 | $0.00014 | $0.00314 |
| Haiku 4.5 | $0.00007 | $0.00157 |
Grade A, and why
review-durability scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 14d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
Flag in a route handler, a WebSocket handler, a watcher tick or anything the server runs: `execSync`/`spawnSync`, `fs.*Sync` on something that can be large or slow (a repo walk, a big file, a network mount), a `git`/`gh` How it starts
The opening of the file, as written. The whole thing — 80 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Durability review
AgentOS is redeployed under running sessions many times a day: every push to main restarts the live server. Sessions, chat turns, tasks and stacks have to come through a restart, a dropped socket or a crash mid-operation without being lost, run twice or wedged. One Node process serves every session, so anything that blocks its event loop freezes all of them. Flag high-confidence problems in code introduced or modified by this branch.
Read first: lib/chat/worker/protocol.ts (the server ↔ worker messages, PROTOCOL_VERSION), lib/chat/worker/client.ts and host.ts, lib/chat/store.ts, lib/db/migrations.ts (the runner) and lib/db/migrations/README.md, and for tasks and stacks lib/stacks/start.ts, restack.ts, land.ts, tick.ts.
Process
- Changed files: use the list in your prompt. If none was given: with a PR number,
gh pr diff <NUMBER> --name-only; otherwisegit diff main...HEAD --name-status -Mplusgit status --short --untracked-files=all. Read new files whole andgit diff main -- <file>for changed ones. - For each changed operation with a side effect, ask: if the server dies right after this line, what does the next start see, and does it finish, retry safely, or do it twice?
- Check every rule below.
- Report.
Rules
The chat worker protocol A chat runs in a worker process (in tmux) that outlives the server; the server reconnects over the Unix socket. Flag:
- A change to
WorkerCommand/WorkerEventthat an older worker or server can't read, without bumpingPROTOCOL_VERSIONor handling both shapes. Workers from an older build are closed only when idle; a running turn is never cut. - A command that isn't idempotent: a send must carry its id and the worker must ignore an id it has seen (
sent), so a send retried after a dropped connection never runs twice. Same for approvals and undo. - Items shown or streamed before they're written to SQLite: the worker saves every item before anyone sees it, and a reconnecting server rebuilds from the store, not from memory.
- State that lives only in the server's memory (a Map of running turns, a pending approval) that a restart loses without a way to rebuild it from the DB or the worker.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 14d ago Changed · +1 tokens per session 8fc058f459d1
- 17d ago Changed 2a5d015142e6
- 22d ago First seen · 80 lines · 69 tokens per session scan A ea2c7ca9b160
review-durability is an agent published in the GitHub repository saadnvd1/agent-os (11 stars, last pushed today), licensed MIT. It adds 70 tokens to every session and 1,570 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 1 finding (runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-10-07.
Other agents, from other repositories
traffic-fidelity-reviewer
Review a cli-web- CLI implementation against its APP.md API map. Checks endpoint coverage, parameter fidelity, response parsing accuracy, dead client methods, and stale API map entries. Returns scored findings. Use during Phase 4 standards review — dispatched by the standards skill.
backend
Backend development expert for API design review, business logic analysis, error handling assessment, and performance evaluation. Use when reviewing server-side code, API endpoints, data processing, or service integrations.
api-reviewer
Reviews API-facing changes for breaking edits and missing docs.
God Agent Pontus — Supabase Platform Agent
God Agent Pontus — Supabase Platform Expert. Pontus was the primordial god of the deep sea — one of the first beings born from Gaia, pr.
God Agent Talos — Web Dev Agent
God Agent Talos — Web Development & Implementation. The bronze automaton Hephaestus built to guard Crete — literally a governed robot that run.
laravel-reviewer
Laravel/PHP code review — Eloquent patterns, security, request validation, queue design.