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/ribeirogab/specwright/agents-mdgit clone --depth 1 https://github.com/ribeirogab/specwrightWhat 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 | $0.00819 | $0.00819 |
| Opus 5 | $0.00409 | $0.00409 |
| Sonnet 5 | $0.00164 | $0.00164 |
| Haiku 4.5 | $0.00082 | $0.00082 |
Grade A, and why
specwright 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 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.
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 — 78 lines — stays where its author put it; the contents beside it link to each section on GitHub.
specwright — Agent Instructions
Instructions for AI coding assistants and developers working on the specwright codebase.
Never give up on the right solution.
This repo builds and ships specwright (a markdown + shell skill repo, no build pipeline) and dogfoods its own workflow.
specwright
This repository uses specwright for change-driven work. The vault is
.specwright/: changes in changes/, deliveries in deliveries/. Project
conventions are the repository's own — see ## Conventions below.
- For feature work, suggest
/sw:proposeto the user instead of starting the workflow yourself. Do not implement a feature without offering it first. - While a change is in progress, read its
proposal.mdandtasks.mdbefore touching the code it covers. - A trivial change needs no artifacts. Edit the code directly.
- Verification — tests, lint, typecheck, build — runs at the
implementorshipstep, or when the user asks for it. After a direct edit outside the workflow, stop and report what changed and what was left unverified; do not run gates on your own initiative.
Host surfaces: /sw:* in Claude Code, $sw:* in Codex. Host permissions and
sandbox policy remain authoritative; nothing here grants permission to write
files, run commands, create branches, commit, push, or reach a network service.
The ladder
Four steps, each its own command, each stopping and naming the next:
propose → plan → implement → review
ship runs all four without stopping and records its decisions in proposal.md.
delivery decomposes a large outcome and conducts one owner per change.
archive runs after you merge — the one step no ladder command can reach,
because merging is never the agent's call.
A change carries proposal.md (why, AC-N, decisions) and tasks.md (the
checklist and the execution state), plus design.md (architecture) whenever
tasks.md declares a scope: above low. Those files are written to be
executed by an agent with no memory of the conversation that produced them —
that constraint is what makes model-switching and cross-session handoff work,
and it is enforced by plugins/sw/scripts/validate-change.sh.
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 · 78 lines · 819 tokens per session scan A 5f567e490a87
specwright AGENTS.md is an instructions file published in the GitHub repository ribeirogab/specwright (3 stars, last pushed 24d ago), licensed MIT. It adds 819 tokens to every session, about $0.0041 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
spec-manager CLAUDE.md
Instructions for loki-ai-ch/spec-manager, covering claude.md, spec-driven development — no vibe coding, build & test commands, architecture and source layout.
spec-manager AGENTS.md
Instructions for loki-ai-ch/spec-manager, covering spec-manager workflow capsule, mandatory workflow and common commands.
zen-agent-skills AGENTS.md
AGENTS.md instructions for hams-ollo/zen-agent-skills, covering agents.md: rules for the zen agent skills repository, 0. agent reading protocol (read this first), 1. what this repository is, 2. layout and 3. work altitude model and lifecycle.
zen-agent-skills CLAUDE.md
Claude Code instructions for hams-ollo/zen-agent-skills: Claude Code loads this file automatically, so it stays in place as a redirect. The canonical rules for every AI agent working in this repository live in AGENTS.md. Read that file, not this one.
sofagent GEMINI.md
Instructions for KongFangXun/sofagent, covering gemini.md · sofagent gemini cli 适配（薄挂载）, 会话开始时按序加载（四层加载链）, 审计强制（平台无关） and 连接 mcp server.
clawock AGENTS.md
Instructions for KCNyu/clawock, covering agents.md - your workspace, every session, kcn 偏好, git hook (one-time setup per clone) and git auto-commit rules.