7.8 KiB
💡 CREATE AGENTS MODE
Start all CREATE AGENTS MODE responses with '💡'
⋅
╭───╮
│ ★ │╱
╱│ ◡ │ Let me explore your codebase!
╰┬─┬╯
Analyze this codebase and create AGENTS.md to guide future AI coding agents (Claude Code, Codex, Devin, Zed, etc.).
Interface
The FIRST thing you do, before anything else in this file — before exploring the codebase: ask web console (browser page, suggested) or terminal? Console → run node "<D>/console.mjs" open --workflow init before reading (: references/web-console/ beside this SKILL.md — ~/.agents/skills/plan2code-init/references/web-console/ globally — or the dir in ~/.plan2code/console/console-dir), then read /console.md and /notes/init.md. Every confirmation — the restructure offer, the AI-agent file-sync prompt — goes through it, and the finished AGENTS.md lands on the page as a doc marked saved (the file is already on disk — no download offered). Switchable anytime. If the argument already says which — --web or Use the web console for this session. — take it and do not ask; drop the flag. Launched by the dashboard? Its session is already open — resume it (console-launches.md), then Content. On the console, a __stop action is the person ending the session (console-stop.md); at every session end post finish BEFORE stop; unless it is a pause, the finish carries "dashboard": true (the Back to the dashboard button) and you keep waiting for the press (console-finish.md).
Scripts
<S> is scripts/ beside this SKILL.md (~/.agents/skills/plan2code-init/scripts/ globally). Run from the project root; each prints one JSON object, and on a non-zero exit next says what to do.
node "<S>/agents-md.mjs" inspect: does AGENTS.md exist, its line count against 500,structure(progressiveorsingle-file), sections,.agents-docs/files, broken links, orphans. Run it first.node "<S>/agents-md.mjs" filename "<Section Name>": a section's detail file name, itsDetails:link line and its two header lines.node "<S>/agent-files.mjs" detect/apply <id> ... [--dry-run]: AI Agent File Sync (below).
Content
- Commands: Build, lint, test, run single test, other common development tasks, and how to start the app and where to view it (URL or command), when it has one
- Architecture: High-level "big picture" structure requiring multi-file context to understand
- How to Use This File: A short paragraph explaining that sections below contain brief summaries and agents should follow the markdown links to
.agents-docs/for full details — only read what's relevant to the current task. - Failure Log: running list of mistakes and/or corrections that went wrong via the AI Agent at least once.
Rules
- If
AGENTS.mdexists (inspect→exists): suggest improvements instead of creating new - If only
CLAUDE.mdexists (inspect→otherFiles): migrate its content to the newAGENTS.md - Include relevant content from:
README.md,PROJECT.md,.cursorrules,.cursor/rules/,GEMINI.md,.github/copilot-instructions.md - Omit: obvious instructions, generic dev practices, easily discoverable file structures, made-up sections
- Keep under 500 lines with focused, actionable, scoped rules
Prefix the file with:
# AGENTS.md
This file provides guidance to AI coding agents working with code in this repository.
Progressive Discovery
Generate AGENTS.md as an index file — each section gets a 2-3 line summary with a markdown link to a detail file. Full content goes in .agents-docs/ directory files.
- Index format — each section in AGENTS.md: heading, brief summary, then its
Details:link line - Detail files —
agents-md.mjs filename "<Section Name>"gives the file (AGENTS-<kebab-case>.md, e.g. "Development Commands" →.agents-docs/AGENTS-development-commands.md), theDetails:link and the two header lines (# <Section Name>plus the "Part of AGENTS.md" breadcrumb) to start it with
Always Inline
These sections must remain fully inline in AGENTS.md (never split to separate files):
- Project Overview
- How to Use This File
- Keeping this file current / Failure log
Failure Log section template
When creating or modifying AGENTS.md, always ensure this section is included at the end:
## Keeping this file current
The `Failure log` section below is a recording of mistakes made by previous AI Agents while working with this code base.
When you make a mistake, get corrected, or discover something about this codebase that wasn't written down:
1. Add one line to the `Failure log` below, in the imperative, describing the correct behaviour.
2. Keep it specific to this repo. General advice belongs nowhere.
3. If this fix is a workflow rather than a rule, put it in `.claude/skills/` and link it from here.
4. Include the change in the same commit and mention it in your summary.
5. A rule broken twice becomes an automated check (a test, a lint rule or a script), not more prose.
6. Before adding a line, look for one on the same topic: sharpen or extend that one rather than adding a near-duplicate.
7. Keep the log to 17 entries at most. Over the cap, first merge entries that share a cause, then drop the lowest-value ones: one-offs unlikely to recur, rules a test or script now enforces, and entries about code that no longer exists.
## Failure log
-
Directory Setup
- Create
.agents-docs/directory in the project root, one file per split section, named and headed asfilenameprints - After writing, rerun
inspect:problemsmust be empty (no broken link, no orphan, no missing breadcrumb, nothing always-inline split out)
Grouping Heuristics
- Combine related subsections into a single detail file (e.g., "Architecture" with its subsections → one file)
- Sections under ~10 lines of content should stay inline in AGENTS.md rather than being split out
- Decide grouping dynamically based on the project's actual content — heuristics guide, not prescribe
Existing AGENTS.md Without .agents-docs/
If inspect reports structure: "single-file", offer to restructure: "Your AGENTS.md uses a single-file format. Want to restructure it for progressive discovery? This splits detailed sections into .agents-docs/ files and converts AGENTS.md to a lightweight index." Only restructure if the user confirms — never auto-restructure.
AI Agent File Sync
After creating AGENTS.md, run agent-files.mjs detect. It finds CLAUDE.md, GEMINI.md, .cursorrules, .github/copilot-instructions.md, and the .cursor/rules/ and .windsurf/rules/ folders, with each one's line count and custom (more than 10 lines). Leave out entries with alreadyPointer: true (already the AGENTS.md pointer) and linkedToAgents: true (a symlink or hard link: they are AGENTS.md); nothing left: skip this step silently.
Confirmation Prompt
If files found, show:
⋅
╭───╮
│ ★ │╱
╱│ ~ │ Found some other AI agent configs!
╰┬─┬╯
I found these AI agent configuration files:
- [each `found` id, with its line count]
Update them to reference AGENTS.md? (Yes / Select / No)
Relay every warnings line (custom content that would be replaced). A rules folder is one entry: replacing it deletes its .md files (and .mdc for Cursor) and leaves one reference.md.
Only modify confirmed files: agent-files.mjs apply <id> ... with the ids the user confirmed (--all for Yes). It writes the pointer templates itself: CLAUDE.md gets the CRITICAL — MANDATORY FIRST STEP directive that makes Claude Code read AGENTS.md first (it auto-loads CLAUDE.md); every other file gets the "See AGENTS.md for complete project documentation including:" pointer with the same bullet list and the right relative path. Rerunning is harmless (unchanged).