agents-control

Take back the control of your coding agents.

What it is: a lightweight, tool-agnostic system that keeps your project's context in a small set of human-readable markdown docs and gives your coding agent (Claude Code, Codex, …) a manifest so it always knows which file holds what — plus skills that route work into the right file. See agents-control.md for the full idea and architecture.

The whole system is 3 files + a folder:

AGENTS.md   ← agent instructions + the doc manifest (the brain)
CLAUDE.md   ← one line, "@AGENTS.md" (Claude adapter; Codex reads AGENTS.md directly)
docs/       ← the actual doc content (PRD, STACK, ROADMAP, DECISIONS, …)
.claude/    ← skills, auto-discovered by the tool

Quickstart

1. Set it up in your repo

From inside the repo you want to bring under agents-control, make the setup skill available to Claude Code, then run it:

# install the one-time setup skill (user-level, available in every repo)
mkdir -p ~/.claude/skills/agents-control
curl -fsSL https://raw.githubusercontent.com/patrickfleith/agents-control/main/install/SKILL.md \
  -o ~/.claude/skills/agents-control/SKILL.md

Then, in Claude Code opened at your target repo, invoke it:

/agents-control

or just ask: "set up agents-control here". The skill will:

• fetch the toolkit from GitHub,
• detect greenfield (empty) vs brownfield (existing code),
• install the doc-maintenance skills into .claude/skills/,
• scaffold the MVP core docs (PRD · STACK · ROADMAP · DECISIONS · TASKS · CHANGELOG),
• generate AGENTS.md (with the manifest) + CLAUDE.md.

It never overwrites existing files silently — it shows what it'll do, does it, then reports.

2. Use it day to day

Your agent now reads AGENTS.md every session and knows where each doc lives. Drive the docs with the bundled skills (ask in plain language or use the slash command):

Skill	What it does
commit	Turn uncommitted work into atomic, conventional commits.
log	Append to STACK / TASKS / IDEAS / CONCERNS / QUESTIONS.
decide	Record and manage decisions in DECISIONS.
glossary	Add and refine canonical terms in GLOSSARY.
write-manual	Write or update the user manual (SUM) from the codebase.

Other doc types (FRD, PLAN, GLOSSARY, SUM, IDEAS, CONCERNS, QUESTIONS, EVALUATIONS, TESTS, UXUI) are created on demand by these skills — you don't scaffold everything upfront.

Development status

Early / MVP. Claude first, Codex later — the shared core (AGENTS.md + manifest + docs/) is portable today; per-tool capabilities are ported as needed.

Doc coverage (✅ = a skill handles file ops · 🚧 = partial, needs a manage skill · 🔴 = not started):

• ✅ SUM (write-manual), DECISIONS (decide), GLOSSARY (glossary), IDEAS / CONCERNS (log)
• 🚧 STACK, TASKS, QUESTIONS — log can append, but a dedicated manage skill is still needed
• 🔴 README, PRD, ROADMAP, FRD, PLAN, EVALUATIONS, TESTS, CHANGELOG, UXUI, DEPLOYMENT — templates exist, no skill yet

Full status list and rationale live in agents-control.md.

License

See LICENSE.