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/heretyc/subagent-mcp/agents-mdgit clone --depth 1 https://github.com/Heretyc/subagent-mcpWhat 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 | $0.04263 | $0.04263 |
| Opus 5 | $0.02132 | $0.02132 |
| Sonnet 5 | $0.00853 | $0.00853 |
| Haiku 4.5 | $0.00426 | $0.00426 |
Grade A, and why
subagent-mcp AGENTS.md 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 2d 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 — 202 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Repository Agent Instructions
PROJECT BOARD LAW - BINDING, NO EXCEPTIONS
https://github.com/users/Heretyc/projects/4 is the canonical Source of Truth for the PLANNING and STATUS of all repo Work. It is subordinate to the alignment chain in Law 2: where Board and Premise conflict, the Premise wins and the Board is corrected.
DEFINITIONS
"Work" = any mutation of git-tracked content: commits, branches, PRs. Wiki edits, releases, tags, and comments are not Work. Board mutations are not Work and never require issues. "Human Review" (HR) = engagement of the human via the structured question tool, presented as if the human has no knowledge of the project or repo. Two modes:
- Decision HR: 2+ options, each with pros and cons.
- Action HR: 1 required action, why it is needed, and the consequence of inaction. HR is NEVER skippable for any authority or reason. One HR session MAY bundle every trigger pending at that moment. "Premise" = The answer to "what is this + why care?": who it's for + problem + what it does + why not alternatives. Canonical copy: first paragraph of README.md, MAXIMUM 1000 characters. "Short Premise" = Verbatim-identical version of Premise (minus the "why not alternatives" aspect) MAXIMUM 350 characters and found in BOTH the Repo and Project Short Descriptions. "Interactive Session" = A harness that has a working structured question tool. Lacking this, the session is considered non-interactive.
DIRECTIVES
- Short Premise must faithfully condense the canonical Premise. Any edit to the Premise updates both Short Premise copies in the same session. If the copies mismatch, truncate, or drift from the canonical: the README is canonical and the mismatch is an HR trigger.
- Code MUST align with repo Spec docs. Spec MUST align with the Premise at all times. The Premise is always the tie-breaker. If in doubt, HR. If the Repo lacks the canonical Premise in README.md, or either Short Description lacks the Short Premise: recon the project specs/code and present a proposed Premise and Short Premise via Decision HR; that same HR authorizes the board issue for the restoration edit (which is Work).
- All Work maps to a board issue. All issues map to Milestones. No unmapped Work.
- Every issue/epic carries ALL required fields at ALL times: Label(s), Priority, Size, Estimate, Iteration, Milestone, Assignee, Relationships, branch/PR link, and updated Status. Satisfiability rules: issues are created fully populated in a single operation, with Iteration taken from the authorizing HR (Law 9, bundled per the HR definition); the branch/PR link is mandatory from the moment the branch or PR exists and MUST be back-filled in the same working session ("none yet" before that; "n/a" for board-only issues); Assignee follows Law 11 (an idle issue may be unassigned or assigned to anyone; an actively-worked issue MUST be assigned to the worker).
- Every PR maps to a fully populated issue.
- Live updates are mandatory: update the mapped item BEFORE, DURING, and AFTER Work. Agents update existing items in real time without seeking permission. While an HR is pending on an item, only Status changes and comments recording the block are permitted on it.
- Every agent MUST fully understand the Board plan and the Premise BEFORE acting. The read-only sweep in Law 13 both requires and satisfies this understanding.
- Any conflict between tasked work and the Board: STOP and deconflict via HR BEFORE any board edit. Conflicts between Board and Premise resolve per Law 2.
- Net-new Work not on the board: HR BEFORE adding it. That HR also supplies the new issue's Iteration and other judgment fields (Law 4).
- Board ops use ONLY the
ghCLI, includinggh apiwhere subcommands lack field coverage (Relationships). Ifghis not authenticated or lacks the project scope: Action HR asking the human to rungh auth refresh -s project(interactive; agents cannot complete it). - Before starting Work on an unassigned issue, assign it to the logged-in
ghuser. If an issue you are tasked to work is assigned to someone else: Action HR to reassign. If the human declines, stand down from that item: you are forbidden from performing Work on any issue not assigned to the logged-inghuser. Idle issues may remain unassigned or assigned to others. - Never store local paths or machine-specific information anywhere on the Board.
- Sweep the Board before starting and after finishing Work: find the most recently COMPLETED "Project Board true-up #". If it completed more than 5 business days ago, or none exists: dispatch 2+ review subagents over all incomplete board items for non-compliance with these Laws (if the harness cannot spawn subagents, perform work directly). HR is mandated on all non-compliant items found (bundle-able). Once the true-up is complete: mark it complete and create the next true-up issue with an incremented #, NO assignee, description = a VERBATIM copy of these Laws from their canonical home (this document at the repo root).
- HR may be DEFERRED only in headless runs where no human is reachable (determined if no structured question tool exists in your harness), and deferral is never resolution. Queue the item as a comment on the dedicated HR-queue board item, set the affected issue's Status to "Awaiting Review" (rename "In Review" to "Awaiting Review" if it exists. If you cannot rename, use "In Review" status), and proceed only with Work unaffected by the pending question; doubt about whether Work is affected resolves to AFFECTED. Any Interactive Session MUST drain the queue (bundled HR) BEFORE starting any new Work. If any queued item is older than 5 business days, ALL Work halts until the queue drains. Every true-up (Law 13) reports the queue's contents. Queued items are never deemed approved, expired, or abandoned.
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.
- 2d ago First seen · 202 lines · 4,263 tokens per session scan A 33bd9bf17f79
subagent-mcp AGENTS.md is an instructions file published in the GitHub repository Heretyc/subagent-mcp (3 stars, last pushed 4d ago), licensed Apache-2.0. It adds 4,263 tokens to every session, about $0.0213 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 instructions, from other repositories
sgrep AGENTS.md
Instructions for XiaoConstantine/sgrep, covering sgrep - agent integration guide, build & test, build, test and install.
agentic-playwright selectors.instructions.md
Instructions for idavidov13/agentic-playwright, covering selector strategy, critical, instructions, phase 1: open and authenticate and phase 2: explore like a user.
hatch3r CLAUDE.md
Instructions for hatch3r/hatch3r, covering hatch3r — development instructions, architecture, development commands, two-axis pillar framework (2.0.0) and orchestrator self-discipline (bypass protection).
heimdall AGENTS.md
Instructions for randomittin/heimdall, covering working in the heimdall repo, the rules that actually block you, environment, conventions and heimdall — this repo is gated.
delivery-loop CLAUDE.md
Instructions for blakemartz/delivery-loop, covering claude.md, what this is, the cardinal rule: everything must stay repo-agnostic, layout and config: the one seam.
claude-for-hk-law CLAUDE.md
Instructions for Buildwise-Studios/claude-for-hk-law, covering claude.md, layout, legal system context, key hong kong ordinances and legal research sources.