Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/sidclawhq/platformnpx agentmods add skills/sidclawhq/platform/sidclaw-governanceWrote 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/sidclawhq/platform/sidclaw-governance)<a href="https://agentmods.dev/skills/sidclawhq/platform/sidclaw-governance"><img src="https://agentmods.dev/badge/skills/sidclawhq/platform/sidclaw-governance.svg" alt="Measured on agentmods" 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 | $0.00059 | $0.01550 |
| Opus 5 | $0.00030 | $0.00775 |
| Sonnet 5 | $0.00012 | $0.00310 |
| Haiku 4.5 | $0.00006 | $0.00155 |
Grade C, and why
sidclaw-governance scanned grade C 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 8d 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.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- run a destructive shell command (`rm -rf`, `DROP TABLE`, `git push --force`, etc.) How it starts
The opening of the file, as written. The whole thing — 166 lines — stays where its author put it; the contents beside it link to each section on GitHub.
SidClaw Governance
You have access to SidClaw — an identity, policy, approval, and audit layer
for AI agents. The governance MCP tools are available as sidclaw_evaluate,
sidclaw_record, sidclaw_approve, sidclaw_policies, sidclaw_session_start,
and sidclaw_session_end.
When to use these tools
Call governance BEFORE executing any action that could:
- modify the file system outside the immediate workspace
- run a destructive shell command (
rm -rf,DROP TABLE,git push --force, etc.) - dispatch a subagent, create a cron, create a team, or trigger a remote action
- hit a network API that sends data to a third party
- deploy code, push to a remote branch, or publish a package
- move money, send messages to real users, or modify medical/legal records
Quick mental model — risk tiers:
| Risk | When | What to do |
|---|---|---|
| < 30 | Local-only, reversible, internal data | Call sidclaw_record after (log-only) |
| 30–70 | Moderate effect, recoverable, might surprise a reviewer | Call sidclaw_evaluate first |
| > 70 | Destructive, irreversible, or touches production / regulated data | Always sidclaw_evaluate first |
Do not bypass governance for convenience. If evaluate returns deny, do not
retry with a different phrasing to try to slip through. Explain the situation
to the user and ask for guidance.
The protocol
Step 1. Start a session (optional but nice for audit grouping)
sidclaw_session_start({
agent_id: "claude-code",
workspace: "/path/to/repo",
branch: "feature/x"
})
→ { session_id: "mcp-sess-..." }
Hold on to session_id for the whole turn.
Step 2. Before each risky action, call sidclaw_evaluate
sidclaw_evaluate({
operation: "bash.destructive",
target_integration: "claude_code",
resource_scope: "rm -rf ./data/",
data_classification: "restricted",
declared_goal: "Clean up stale migration artifacts per user request",
systems_touched: ["filesystem"],
reversible: false,
risk_score: 95
})
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.
- 8d ago First seen · 166 lines · 59 tokens per session scan C 1c3048604125
sidclaw-governance is a skill published in the GitHub repository sidclawhq/platform (14 stars, last pushed today), licensed Apache-2.0. It adds 59 tokens to every session and 1,550 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
ziran
ZIRAN is an open-source security testing framework for AI agents. It discovers dangerous tool chain compositions via knowledge graph analysis, detects execution-level side effects (not just text output), and runs multi-phase trust exploitation campaigns that model real attacker behavior.
docs
ACMI is a universal architectural framework for giving AI agents persistent, real-time context. It replaces fragmented, multi-table database joins with a single, lightning-fast Key-Value engine (Upstash Redis) optimized specifically for LLM context windows.
brain-save
Saves a single fact, decision, pattern, or convention into your governed knowledge brain so it can be recalled later — and retires memories that are outdated. Side-effecting: it writes to your durable corpus, so it never auto-fires — invoke it explicitly. Use when you want the brain to remember something specific…
brain
Answers questions about your own systems, notes, decisions, runbooks, and conventions from your governed knowledge brain, returning a qmd:// citation for every claim — receipts, not recall. Use when you want to know what your brain has captured about your own architecture, infrastructure, decisions, or conventions…
aura
Give this agent a real email address, a webhook URL, durable memory across runs, and the ability to wait for an event without burning tokens. Use when a task needs a verification code sent by email, needs to wait for a webhook or an approval, needs to remember something after the session ends, or needs to survive a…
octopoda-memory
Persistent memory across sessions — recall, store, share, audit decisions, snapshots, and version history.