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 instructions/faxik/codebugs/claude-mdgit clone --depth 1 https://github.com/faxik/codebugsWhat 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 | $0.45188 | $0.45188 |
| Opus 5 | $0.22594 | $0.22594 |
| Sonnet 5 | $0.09038 | $0.09038 |
| Haiku 4.5 | $0.04519 | $0.04519 |
Grade C, and why
codebugs CLAUDE.md scanned grade C 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 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.
Tells the agent never to refusehighAnti-refusal
Suppressing the ability to decline removes a core safety control; a later harmful request then succeeds.
**(12) STRIP WITH VISIBILITY, ON THE ADD PATH ONLY (CB-56/CB-60, closed by the wire pin at CB-160).** `add`/`batch_add` do not REFUSE a caller-supplied `meta` key that is identity machinery's own OUTPUT (`occurrences`, ` Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
over stdio — and `subprocess` is used legitimately for git, which can of course run `curl`. What is Runs shell commandslowCapability
Expected in a hook, worth knowing in a rule or an instructions file.
- **`OSError` arrives from ambient sources, not just from `open()` — and a guard spelled as one errno is an enumeration (CB-79).** A sweep for `open(` structurally cannot see `os.getcwd()`, `subprocess`, `Path.read_text` How it starts
The opening of the file, as written. The whole thing — 1,156 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Codebugs
AI-native code finding & requirements tracker. SQLite-backed, exposed via MCP server + CLI.
Workflow — main is never edited directly
Every code edit happens on a short-lived branch, in a worktree, and tools/ enforces it.
Borrowed from ../autosorter (2026-08-16), including a scaled-down port of its
tools/worktree-*.sh harness.
This section once said the harness was unnecessary — "plain git is enough" — and that claim was
falsified within two hours by the very rule it was introducing (CB-50). It landed at 13:37 on
2026-08-16 (2957070) mandating a typed branch and git merge --no-ff; at 15:30 main was advanced
by merge worktree-cb-45-similarity-seam: Fast-forward — no type prefix, no merge commit, and by
then pointing at main's own SHA, so every further merge would fast-forward again. Two sentences
earlier the same section had stated the reason — a convention that exists only as a pattern in the
log is not a rule — and then declined to bind it. Prose cannot enforce prose; the harness below
is the fix.
What is now mechanically enforced (tools/install-hooks.sh arms it; run once per clone):
| Rule | Mechanism | Refuses with |
|---|---|---|
Branch carries fix/|feature/|refactor/|docs/ |
_guard_branch_type (7) + pre-commit hook (1) |
exit 7 / 1 |
Nothing but .claude/plans/*.md or .claude/plans/briefs/*.html is committed on main |
pre-commit hook | exit 1 |
| A plan note committed on main is NAMED in the commit message | commit-msg hook | exit 1 |
A cascade id added to .claude/plans/CASCADE-IDS.md on main is the one tools/cascade-mint.sh would have computed (max+1 per family, annulled lines and mentions included) |
pre-commit hook | exit 1 |
A merge onto main comes from a typed local branch, or from main's own upstream main |
pre-merge-commit hook (clean merge) + pre-commit hook (conflicted merge) | exit 1 |
| An in-progress cherry-pick/revert marker no longer exempts a commit | pre-commit hook | exit 1 |
| Integration never fast-forwards | --no-ff + git config merge.ff false |
— |
| One integration at a time | flock on .worktrees/.integrate.lock |
exit 1 |
| The tested state still matched at the moment it was re-checked | in-lock SHA re-check | exit 13 |
| The suite ran under the interpreter main has | _guard_interpreter_matches_main |
exit 14 |
| This clone is actually armed | _guard_enforcement_armed |
exit 12 |
| Main has main checked out, and is clean | _guard_workspace_on_main, _guard_main_clean |
exit 8, 11 |
| The branch actually carries a change | _guard_nonempty_diff |
exit 9 |
| No conflict markers, no scratch or temp file at root, no stale base | _guard_conflict_markers, _guard_untracked_scratch_at_root, _guard_stale_base |
exit 5, 4, 6 |
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 · 1,156 lines · 45,188 tokens per session scan C 521fb95fb9f9
codebugs CLAUDE.md is an instructions file published in the GitHub repository faxik/codebugs (1 stars, last pushed yesterday), licensed MIT. It adds 45,188 tokens to every session, about $0.2259 per session on Opus 5. A static security scan graded it C with 3 findings (tells the agent never to refuse, makes network calls, runs shell commands). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other instructions, from other repositories
ai-workflow AGENTS.md
Instructions for cunhaax/ai-workflow, covering ai workflow template, rules — non-negotiable, project overview, commands and architecture.
trackly-cli CLAUDE.md
Claude Code instructions for trackly-app/trackly-cli, covering trackly-cli, tech stack, backend production source of truth, directory structure and key commands.
com-example-config-AsyncTestConfig
Instructions for PIsberg/vibetags, covering copilot instructions for asynctestconfig and immutable type.
pupila CLAUDE.md
Instructions for FranRom/pupila, covering claude.md, where things live, stack, run locally and repo layout.
claude-bridge AGENTS.md
Instructions for kadaba/claude-bridge, covering agents.md — using claude-bridge from any ai coding cli, two independent channels, 1. coordination (the conversation), 2. file transfer (the bytes) — same for every tool and workflows.
UsageAtlas AGENTS.md
Instructions for SRX9/UsageAtlas, covering repository guidelines and coding approach.