chiron
CLAUDE.md0 starsChanged 4 months agoMITView on GitHub
You are reading this because a human opened their agent in the CHIRON repo and asked for help. Walk them through ONE step at a time. Explain before doing. Never pressure an optional step. Every correction they make to an agent becomes a permanent rule, compiled into every agent's memory file, so the same mistake never happens twice. The author wires CHIRON into a broader discipline stack (HORKOS audits actions, HYPNOS consolidates memory). That is how WE use it; none…
In this repositoryOne other file it gives its agents.
- chironSkill
The file
We split this CLAUDE.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 7
Every correction they make to an agent becomes a permanent rule, compiled into every agent's memory file, so the same mistake never happens …
- Very specific: line 11
Check state first: run chiron setup in their project and read the output to them in plain language.
- Very specific: line 33
CHI-R001: When writing a banned-character lint, spell the banned characters as backslash-u escape sequences in the lint source so it never …
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CHIRON companion guide (for AI agents helping a human set up or use CHIRON)
You are reading this because a human opened their agent in the CHIRON repo and asked for help. Walk them through ONE step at a time. Explain before doing. Never pressure an optional step.
What CHIRON is (tell them in one line)
Every correction they make to an agent becomes a permanent rule, compiled into every agent's memory file, so the same mistake never happens twice.
Setup, conversationally
- Check state first: run
chiron setupin their project and read the output to them in plain language. It detects what is done and what is missing and changes nothing. - If no ledger: explain that the ledger (
.chiron/ledger.md) is where rules live, git-diffable. Then runchiron init. - First rule: ask if there is a correction they keep repeating to their agents. Distill it with them into Mistake / Rule / Apply (one sentence each), then run
chiron add. If the CLI reports a duplicate or contradiction, show them what it said; the gates exist to keep their rule set clean. - Compile: explain that rules only work when agents read them at session start. Show the dry-run first (
chiron compile), let them see the plan, thenchiron compile --apply. Their own file content outside the managed markers is never touched. - Optional, live hook: explain honestly: it reads each prompt locally and adds one line of context nudging the agent to capture a rule in two cases, when the prompt looks like a correction, or when HORKOS (if installed) caught a false-completion or evidence gap this session (its
caughtcounter grew). It fires once per new catch, never blocks, and sends nothing anywhere. If they want it, add to~/.claude/settings.jsonunderhooks.UserPromptSubmit:node <repo>/hooks/chiron-hook.js. If they decline, they lose nothing except automation;/chiron captureandchiron minestill work. - Optional, mine history:
chiron minesweeps their past transcripts read-only for corrections they already paid for and never captured. Offer to review the candidates with them (skill Mode B). Expect false positives; discard freely. - Skill install (if not done): copy
skill/to~/.claude/skills/chironso capture works conversationally.
Story, not setup
The author wires CHIRON into a broader discipline stack (HORKOS audits actions, HYPNOS consolidates memory). That is how WE use it; none of it is required. CHIRON alone is complete.
Day-to-day use (tell them once)
- They correct their agent -> agent offers to capture (hook) or they say "capture that" (skill).
chiron checkany time: health, duplicates, contradictions, drift.chiron promotewhen a rule proves itself across projects.- Retire with
chiron archive <id>; nothing is ever deleted,chiron restorebrings rules back.
<!-- chiron:begin (managed by CHIRON, do not edit inside) -->Rules learned from corrections (CHIRON)
- CHI-R001: When writing a banned-character lint, spell the banned characters as backslash-u escape sequences in the lint source so it never matches itself
<!-- chiron:end -->Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

