Install
$ agentstack add skill-drag88-claude-dev-framework-agentsmd-generator ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
AGENTS.md Generator Skill
Auto-generate AGENTS.generated.md from project rules. This is the Codex / GPT-5.5 counterpart to CLAUDE.md. Also picked up by Cursor, Aider, Jules, Devin, Zed, GitHub Copilot, and other coding agents that follow the AGENTS.md convention.
When to Activate
.claude/rules/exists but noAGENTS.mdorAGENTS.generated.mdin project root- User asks "generate agents.md" or similar
- User asks "create Codex docs" or "set up AGENTS.md"
- After
/cdf:rules generatecompletes (auto-chain alongside claudemd) - After
/cdf:rules claudemdcompletes when AGENTS.md is also missing (sibling auto-chain)
What It Does
Executes the /cdf:rules agentsmd behavioral flow. The command definition in commands/rules.md is the single source of truth for:
- What gets extracted from which rule files
- Host-specific translation rules (strip
@fileimports, replace XML tags with prose, neutralize Claude-only tool names) - The output template and structure
- Line budget and guidelines
Design Principles (Codex-aligned, from rules-templates/agentsmd-codex-rulebook.md)
- Single source of truth, two host-tailored outputs. Same
.claude/rules/feeds bothCLAUDE.mdandAGENTS.md. Translate framing per host, not facts. - Target 40-60 lines. Codex caps the combined
AGENTS.mdchain at 32 KiB (project_doc_max_bytes), so root AGENTS.md stays slim and pushes per-package detail into nestedAGENTS.md. - No
@fileimports. Codex parses Markdown literally —@README.mdreads as the literal string, not a transclusion. Inline a 2-line summary if you need the content. - **No `
blocks.** XML processing tags (,`) are Claude-specific. Codex sees the angle brackets as text. - Neutral tool vocabulary. Avoid
Task tool,TeamCreate,Agent tool,subagent_type. Usesubagent,parallel investigation,host skill. - Nested
AGENTS.mdreplaces path-scoped rules. Codex does not honorpaths:frontmatter. Polyglot repos push per-package conventions into/AGENTS.md. AGENTS.override.mdis the overlay convention. When a sibling agent ecosystem (Cursor, Aider) needs different framing, ship it asAGENTS.override.mdrather than forking the root file.- Every line passes: "Would removing this cause Codex to make mistakes?" If not, cut it.
What Belongs Where
| Content | Location | Why | |---------|----------|-----| | Role (one sentence) | AGENTS.md | Anchors tone and scope cheaply | | Overview, Quick Start, Critical Rules | AGENTS.md | Essential context for every interaction | | Tool and subagent policy (prose only) | AGENTS.md | Codex needs explicit parallel-work authorization without XML | | CDF tools available (compressed routing table) | AGENTS.md | Without explicit routing, Codex falls back to generic approaches | | Commit message conventions | AGENTS.md | 1-line pointer, high-frequency need | | Key directories | AGENTS.md | Orientation for every task | | Project-specific gotchas | AGENTS.md | Non-obvious things any model would get wrong | | @file imports | NOT in AGENTS.md | Codex parses literally | | ` blocks | NOT in AGENTS.md | No host semantics in Codex | | Claude-specific tool names | NOT in AGENTS.md | Confuses Codex users | | Full workflow detail | .claude/rules/workflow.md + nested AGENTS.md | Codex picks up per-directory context naturally | | Architecture, patterns, tech stack | .claude/rules/*.md | Reference material; do not re-inline in AGENTS.md | | Per-package conventions (polyglot repos) | /AGENTS.md` | Nested AGENTS.md replaces path-scoped rules |
Authoritative Rulebook (loaded at generation time)
Step 0 of every generation: Read rules-templates/agentsmd-codex-rulebook.md (vendored into CDF). Treat it as the single source of truth for Codex AGENTS.md authoring, including the four core differences from CLAUDE.md (no @file imports, no XML-tag semantics, nested AGENTS.md replaces path-scoped rules, AGENTS.override.md overlay convention). The Design Principles above become a fallback baseline only if the rulebook is missing.
Updating the rulebook: Replace rules-templates/agentsmd-codex-rulebook.md with the latest version when Codex / OpenAI publishes new guidance, and commit. The CDF version bump tracks the rulebook update.
Generation flow:
- Read
rules-templates/agentsmd-codex-rulebook.mdas the rulebook. If missing, fall back to the embedded principles above. - Read the project's
.claude/rules/files to extract overview, tech stack, quick-start commands, key directories, project-specific gotchas. - Translate host-specific framing: strip
@fileimports, replace XML tags with prose, neutralize Claude-only tool names. - Construct each section of
AGENTS.generated.mdagainst the rulebook. Stay under 60 lines. - Final pass: walk the rulebook's audit checklist. Fix every "no" before writing.
- Write the file.
- For polyglot repos (multiple
package.json/pyproject.toml/go.mod), also recommend nestedAGENTS.mdfiles per package and list the candidate paths.
Related Commands
/cdf:rules agentsmd— manually trigger generation (behavioral spec lives here)/cdf:rules generate— regenerate rules (auto-chains to claudemd + agentsmd)/cdf:rules(no subcommand) — full chain: generate rules, then both host files
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: drag88
- Source: drag88/claude-dev-framework
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.