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/toolate28/spiralsafe/coherence_protocolgit clone --depth 1 https://github.com/toolate28/SpiralSafeWhat 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 | $0.00000 | $0.04766 |
| Opus 5 | $0.00000 | $0.02383 |
| Sonnet 5 | $0.00000 | $0.00953 |
| Haiku 4.5 | $0.00000 | $0.00477 |
Grade B, and why
COHERENCE_PROTOCOL scanned grade B with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
atom_tag=$(cat .claude/last_atom) The source is not reproduced here
No licence file
A repository with no LICENSE is all rights reserved by default, so the body is not copied here. The metadata, the measurements and the link are.
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.
- yesterday First seen · 802 lines · 0 tokens per session scan B 23a5da160f12
COHERENCE_PROTOCOL is an agent published in the GitHub repository toolate28/SpiralSafe (10 stars, last pushed 17d ago), with no licence file. It costs nothing until one of its globs matches a file; then it loads 4,766 tokens. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
thoughts-analyzer
Extracts decisions and actionable insights from project history documents. Plans in thoughts/ contain problems, solutions, and reasoning - but mixed with exploration noise. Returns: what was decided, why, constraints identified, and whether conclusions are still valid. Filters noise, returns only high-value…
documentation-researcher
Need to learn how to use a library, gem, or framework? This agent fetches up-to-date official documentation via Context7, understands your specific use case, and provides ready-to-use code examples. Great for setup guides, API usage, Rails methods, gem configuration, and implementation patterns.
swarm-worker
Execute a single swarm task in an isolated git worktree. Receive a fully-specified recipe from the main context — a scenario recipe plus an implementation contract — then run Skill(scenario) followed by Skill(implement) and report JSON status. Make no design decisions and do not expand scope. Invoked exclusively by…
review-rails
Rails conventions and architecture reviewer for PR audits. Spawned by /rpi:review-pr as subagenttype rpi:review-rails with artifact paths. Ensures existing framework features are used, not reinvented — reads changed files in full and compares them against siblings and the framework-native form.
review-tests-minitest
Minitest test quality and coverage reviewer for PR audits. Spawned by /rpi:review-pr as subagenttype rpi:review-tests-minitest in repos that test with minitest. Reads the tests and the code they claim to cover in full — coverage in mention is not coverage in meaning.
review-docs
Documentation reviewer for PR audits. Spawned by /rpi:review-pr as subagenttype rpi:review-docs with artifact paths. Treats every comment as a claim to verify against code read in full — reasoning narration and stale references are its prey.