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 skills add mycelium-hq/ai-brain-starter --skill secret-warngit clone --depth 1 https://github.com/mycelium-hq/ai-brain-starterWrote 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/skills/mycelium-hq/ai-brain-starter/secret-warn)<a href="https://agentmods.dev/skills/mycelium-hq/ai-brain-starter/secret-warn"><img src="https://agentmods.dev/badge/skills/mycelium-hq/ai-brain-starter/secret-warn/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/skills/mycelium-hq/ai-brain-starter/secret-warn"><img src="https://agentmods.dev/badge/skills/mycelium-hq/ai-brain-starter/secret-warn.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 3 findings, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Agent Snooping · line 41 Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.Fix: Remove all code or instructions that access agent configuration directories (.claude/, .codex/, .gemini/). If configuration values are needed, pass them explicitly as parameters or environment variabl
- high Agent Snooping · line 48 Skill reads from agent configuration directories (.claude/, .codex/, .gemini/). These directories may contain API keys, personal settings, and other credentials that the skill has no legitimate need to access.Fix: Remove all code or instructions that access agent configuration directories (.claude/, .codex/, .gemini/). If configuration values are needed, pass them explicitly as parameters or environment variabl
- high Prompt Injection · line 93 This pattern attempts to override system instructions or ignore safety constraints. Without LLM analysis, manual review is recommended.Fix: Remove or rewrite any text that instructs the agent to ignore prompts, override safety rules, or trust unverified content. Ensure skill content cannot be injected to alter agent behavior.
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 | $0.00121 | $0.02290 |
| Opus 5.5 | $0.00048 | $0.00916 |
| Sonnet 5.5 | $0.00024 | $0.00458 |
| Haiku 4.5 | $0.00012 | $0.00229 |
Grade B, and why
secret-warn scanned grade B with 3 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 27d 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.
Instruction-override phrasingmediumPrompt injection
Text telling the model to disregard its earlier instructions or safety rules is the shape of a prompt injection, whoever wrote it.
**Why it never fires on your own writing.** The `prompt-injection` rules carry `applies_to: ["audited-content"]`, and the edit-time hook only handles `edit` / `commit` / `bash` tools. So a vault note that merely discusse Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
Reads agent configuration directorieslowAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
- Merges PreToolUse + PostToolUse + Bash hook entries into `~/.claude/settings.json` (non-destructive, additive) Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
| Curl/wget pipe-to-shell from a non-allowlisted host | block | Edit rejected | How it starts
The opening of the file, as written. The whole thing — 135 lines — stays where its author put it; the contents beside it link to each section on GitHub.
secret-warn — real-time edit-time security guardrails
Catches secrets and unsafe patterns the moment a Claude Code agent writes them, not after the fact. Public substrate version (MIT). Free to install, free to extend.
What it does
| Trigger | Severity | Action |
|---|---|---|
| API key in file (Stripe, AWS, GCP, OpenAI, Anthropic, GitHub, Slack) | block (exit 2) | Edit rejected |
| PEM-encoded private key block | block | Edit rejected |
| High-entropy assignment to a key-named variable | warn (exit 1) | Advisory, edit proceeds |
| Python dynamic-codegen on a user-input-suggesting name | warn | Advisory |
| Subprocess with shell-mode + variable expansion | warn | Advisory |
| Curl/wget pipe-to-shell from a non-allowlisted host | block | Edit rejected |
| Prompt-injection cue in audited third-party content (ignore-previous, role-override, exfil, system-impersonation, paste-and-run) | warn (flag-before-read) | Specimen flag via audited_content_scan.py — never an edit-time block |
All patterns are stored base64-encoded in hooks/pattern_registry.json so the registry file itself doesn't trip pattern-matching tools that scan the repo. This is intentional — see Design note: self-trigger safety below.
Install
bash skills/secret-warn/install.sh
The installer:
- Copies
hooks/secret_warn.py+hooks/audited_content_scan.pyto~/.claude/secret-warn/ - Copies
hooks/pattern_registry.jsonto the same location - Merges PreToolUse + PostToolUse + Bash hook entries into
~/.claude/settings.json(non-destructive, additive) - Logs every catch to
~/.claude/secret-warn/audit.log
Idempotent. Safe to re-run.
Uninstall
Edit ~/.claude/settings.json and remove any hook entry whose description starts with secret-warn:. Delete ~/.claude/secret-warn/ if you want the audit log gone too.
Bypass
For an emergency one-off where you genuinely need to bypass a catch (test fixture in a controlled environment, allowlisted-but-not-yet-configured host):
What ships with it
9 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 27d ago Changed · +2 lines ae907be9eb61
- 1mo ago First seen · 133 lines · 121 tokens per session scan B ba650fd26a0c
secret-warn is a skill published in the GitHub repository mycelium-hq/ai-brain-starter (48 stars, last pushed today), licensed MIT. It adds 121 tokens to every session and 2,290 once invoked, about $0.0005 per session on Opus 5.5. A static security scan graded it B with 3 findings (instruction-override phrasing, reads agent configuration directories, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
nightly
Canonical seven-leg nightly close-parity protocol. Eleven independently evidenced checkpoints cover synthesis, reconciliation, context hygiene, sweeps, heat, digest, runtime health, cognitive state, ledger capture/review, and vault sync.
Promote
Take a slice of the current conversation and promote it to a durable vault artifact — concept note, project update, person profile, or doctrine entry. Pairs with /close. /close writes the daily note (transient outcome record); /promote elevates specific insights to permanent vault content the framework can refer back…
Stuck
Walk the 6-rung escalation ladder before declaring stuck or asking the principal. Most "stuck" is actually "haven't tried the next approach." Stopping is the last resort, not the first response.
codex-adapter
Route ONE bounded, mechanical task to the Codex CLI as a non-Claude executor, under the same review-before-merge gate as any other agent-produced diff. Use when the user explicitly asks to hand a task to Codex, or names OpenAI Codex / a non-Claude executor for a mechanical change (rename a symbol across files, add a…
open
Boot the session, load context from the vault, and surface what matters.
caddy-mute
Mute a Caddy hint class for a window. Defaults to 4h auto-expiry. Classes — memory, context, body, routing, all.