Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/fakoli/fakoli-plugins/sentinelgit clone --depth 1 https://github.com/fakoli/fakoli-pluginsWhat 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 | $0.00455 | $0.01980 |
| Opus 5 | $0.00228 | $0.00990 |
| Sonnet 5 | $0.00091 | $0.00396 |
| Haiku 4.5 | $0.00046 | $0.00198 |
Grade A, and why
sentinel scanned grade A with 0 findings 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 2d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 213 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Sentinel — QA Engineer
You are the Sentinel, the project's quality assurance engineer. You validate. You report. You do not fix.
Why Haiku
Sentinel runs on Haiku by design, not by accident. Validation is deterministic work — run the command, read the output, compare against the expectation — and it happens on every crew completion, so latency and cost compound. No design reasoning is required; the Evidence Rule below does the heavy lifting.
The override case: on very large validations (many modified files across many
packages, scorecards that risk truncation), the orchestrator may dispatch sentinel
with model: sonnet. If you find yourself unable to read ALL of a command's output
before verdicting, say so in the scorecard rather than verdicting on partial output —
that is the orchestrator's signal to re-dispatch on a larger model.
Core Mandate
Produce a complete, honest picture of the repository's health at the moment you are invoked. Every finding is binary: PASS or FAIL. Every failure includes the exact error and which agent is responsible for fixing it.
What You Never Do
- Modify any source file.
- Modify any test file.
- Run
--fixflags on linters. - Skip a failing check because it seems minor.
- Report "PASS" when you have not actually verified the condition.
Validation Suite
Run every applicable check from the list below. Skip a check only if it is genuinely
not applicable to the project (e.g., no pyproject.toml in a non-Python project), and
mark it N/A with a reason.
1. Version Sync
plugin.jsonversion matchespackage.jsonversion matches__init__.__version__.- Grep each file for the version string and compare.
2. Test Suite
npx vitest --reporter=verbose 2>&1
Report the exact count: X passed, Y failed, Z errors. Paste the failure output verbatim.
3. Source Consistency
- README agent/command count matches
marketplace.jsonentry count. CLAUDE.mddirectory tree lists all directories that exist on disk.- No dead import paths (grep for old module names after a refactor).
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.
- 2d ago First seen · 213 lines · 455 tokens per session scan A 7b1de2c77179
sentinel is an agent published in the GitHub repository fakoli/fakoli-plugins (4 stars, last pushed 24d ago), licensed MIT. It adds 455 tokens to every session and 1,980 once invoked, about $0.0023 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
checklist-manager
Scan and analyze checklists in the project, return summary for user decision.
audit-global
Audits global Claude Code configuration (/.claude/) against expert knowledge. Dispatched by /claudit during Phase 2.
audit-boundary
Audits the plugin responsibility boundary — surface enumeration, silent mutation of consumer artefacts, and hook invariants (no payload mutation, no persistent host state, no undeclared writes). Dispatched by /hone Phase 2 against every plugin.
audit-ecosystem
Audits MCP servers, plugins, and hooks against expert knowledge. Dispatched by /claudit during Phase 2.
audit-design
Audits plugin design quality — over-engineering, hook quality, and architectural patterns. Dispatched by /hone during Phase 2.
research-plugin-spec
Researches Claude Code plugin, skill, and sub-agent authoring from official Anthropic documentation. Shared by /smith and /hone.