| FMRS | 61 / 100 · C | 79 / 100 · B |
|---|
| Reliability | 8 / 20 | 12 / 20 |
|---|
| Security and permissions | 13 / 20 | 18 / 20 |
|---|
| Maintenance | 13 / 20 | 16 / 20 |
|---|
| Documentation | 14 / 20 | 18 / 20 |
|---|
| Setup experience | 13 / 20 | 15 / 20 |
|---|
| Best for | - Developers running AI coding agents in production or high-blast-radius workflows who need an external enforcement boundary
- Engineering/security/platform teams wanting auditable, inspectable gate decisions
- Teams converting repeated corrections into reusable rules instead of relying on model memory
| - Engineering teams scaling AI coding agents while controlling spend
- Platform teams that need auditable, rollback-aware agent execution records
- Organizations using multiple agents (Claude, Codex, Gemini) under one governed flow
|
|---|
| Not for | - Small teams happy with a small static hand-written denylist (native hooks suffice)
- Those expecting gating to improve model generation quality (it intercepts execution only)
- Cursor/Cline/OpenCode users handling irreversible actions, since advisory verdicts can be ignored
| - Developers who just want a simple code-completion helper without run governance
- Teams unwilling to accept local JSONL run records with HMAC signing
- Environments without a supported adapter (Claude/Codex/Gemini CLIs, direct provider, or verifier-only)
|
|---|
| Required permissions | - Runs locally over stdio; reads/writes ~/.claude or in-repo .claude config, hooks, and lesson stores
- Writes local state (.thumbgate/), JSONL logs, and task-outcome receipts
- Requires modifying the agent's MCP and PreToolUse hook configuration files
- Task-outcome decisions need THUMBGATE_HUMAN_REVIEWER_ID and an independently revocable human reviewer key
| - Read/write access to the configured repo root (writable scope restrictable via --allow-path/--deny-path)
- Ability to invoke the chosen agent CLI (Claude/Codex/Gemini, etc.) and its underlying model account
- Local disk writes for run records, receipts, and share outputs (share/run-receipt., etc.)
- Execution of user-configured verifier commands (e.g. npm test)
|
|---|
| Risks and side effects | - Advisory integrations (Cursor, Cline, OpenCode, Amp) can be ignored by the agent — no true blocking of irreversible actions
- In strict mode, false positives from stale or noisy rules can block legitimate work; recovery relies on break-glass
- Ships warn-by-default: most high-risk commands are not blocked unless strict enforcement is on
- Dashboard token/dollar savings are estimates, not measured provider usage
- Lesson DBs contain sensitive engineering context and must stay gitignored in per-project installs
- Regulated-industry templates are roadmap directions, not shipped compliance capabilities
| - Verifier commands are executed; poorly chosen verifier commands carry security risk despite preflight policy checks
- Cost and token figures may be estimates (labeled with provenance) and should not be treated as settled accounting
- Receipt integrity must be verified before use as trustworthy evidence; missing rollback evidence is flagged as EVIDENCE_BOUNDARY
- Policy defaults come from martin.config.yaml and can be overridden by CLI flags; misconfiguration can loosen safety boundaries
|
|---|
| Supported clients | Claude Code, Claude Desktop, Cursor, Codex, Gemini CLI, Cline, OpenCode, Amp | Claude Code, Codex, Gemini CLI, Cursor, VS Code, GitHub Copilot, Continue |
|---|
| Tools | 13 | 0 |
|---|