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 skills/jayofemi/claude-harness-forge/forge-rulesnpx skills add JayOfemi/claude-harness-forge --skill forge-rulesgit clone --depth 1 https://github.com/JayOfemi/claude-harness-forgeWrote 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/jayofemi/claude-harness-forge/forge-rules)<a href="https://agentmods.dev/skills/jayofemi/claude-harness-forge/forge-rules"><img src="https://agentmods.dev/badge/skills/jayofemi/claude-harness-forge/forge-rules.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 | $0.00096 | $0.00978 |
| Opus 5 | $0.00048 | $0.00489 |
| Sonnet 5 | $0.00019 | $0.00196 |
| Haiku 4.5 | $0.00010 | $0.00098 |
Grade A, and why
forge-rules scanned grade A with 0 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 4d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 49 lines — stays where its author put it; the contents beside it link to each section on GitHub.
forge-rules
Turn the rule surfaces only the owner can decide into a short interview. You hold the pen; the owner supplies every rule. The output bar: each surface the owner chose to fill carries exactly what they stated, read back to them, and nothing else in the workspace changed.
Ground rules
- Anchor at the workspace root first. Every file below lives at the root, the folder whose
CLAUDE.mdis the constitution (it also holdshooks/,STANDARDS/, andtools/). If the session is open elsewhere (a project folder), resolve every path against that root and say so. If you cannot find the root, stop and ask for it; never create these files anywhere else. - Never invent a rule. Every line written comes from the owner's own words in this conversation (tightened for clarity is fine; read the tightened form back).
- One surface at a time, one short plain-words question each. Skipping any surface is fine; the interview is re-runnable and picks up from whatever the files already hold.
- Echo every write. After writing a file, show what it now says in a line or two.
- Touch ONLY the files named below, plus
hooks/gated-tools.txtwhen surface 1 calls for it. Project repos are never touched.
The surfaces, in paying-off order
1. Hard lines -> hooks/hard-lines.txt
Ask: which git operations must an agent never do on its own? Think pushes (which repos, if any), deploys, tags, history rewrites, anything money-touching. Replace the file's ENTIRE content with the answer, stated as short verbatim lines. The git gate quotes this file in every block message; until it is filled, the gate blocks gated operations and says it is waiting. If they name a deploy or publish tool, check hooks/gated-tools.txt for it and offer to add it there so the gate enforces what the line says.
2. Never-publish list -> deny-list.txt at the root
Ask: which names must never appear in anything published from this workspace? Their name and employer, internal project names, private domains, personal paths. Create the file if it is missing (one entry per line); append if it exists, never dropping an existing entry. Note once that the sweep (tools/deny-sweep.mjs) checks against this list before anything goes public, and that the list itself is private data.
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.
- 4d ago First seen · 49 lines · 96 tokens per session scan A 4bf8c349b7a4
forge-rules is a skill published in the GitHub repository JayOfemi/claude-harness-forge (2 stars, last pushed 1mo ago), licensed MIT. It adds 96 tokens to every session and 978 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
list-recent-decisions
Surface the user's recent AxonFlow governance decisions — answers "what just got blocked", "show me my recent denials", or feeds a decision-history forensic flow.
deep-research-sprint
Orchestrates the full research sprint: fetch-before-act source warming, Research Scout → Synthesizer → Reviewer → Archivist → Orchestrator commit pipeline. USE FOR: starting a research topic with a formal synthesis deliverable; coordinating the research fleet end-to-end; producing D4 docs/research/.md files with YAML…
session-management
Governs the full agent session lifecycle: scratchpad init/close, encoding checkpoint at session start, compaction guard, phase gate protocol, and session summary. USE FOR: starting a new session (prunescratchpad.py --init); writing ## Session Start with governing axiom; managing cross-agent context in .tmp/ / .md…
agent-file-authoring
Encodes the conventions for authoring .agent.md files in .github/agents/: YAML frontmatter requirements, required section headings, relative path rules, cross-reference density, and CI validation. USE FOR: creating or modifying .agent.md role files (VS Code: Custom Agents); diagnosing validateagentfiles.py CI…
prompt-archaeology
Defines the prompt archaeology ritual: systematically mining prior session scratchpads, commit messages, and issue comments to surface decision rationale, failed approaches, and discovered constraints that should be encoded into the substrate. USE FOR: post-sprint review of older session artifacts (.tmp/ scratchpads…
conventional-commit
Enforces Conventional Commits format for the EndogenAI Workflows repository. USE FOR: composing commit messages (type(scope): description format); reviewing commit message format before push; understanding allowed types (feat/fix/docs/chore/test/refactor/ci/perf) and scopes…