ai-native-sdlc-playbook
CLAUDE.md0 starsChanged 2 months agoMITView on GitHub
Companion to Anthropic's "The AI-Native SDLC playbook" (Louis Claxton, 2026-08-21). Three layers: the article's files verbatim (hash-locked), the pieces the article describes in prose implemented, and a plugin plus installer that place them into another repo. Nothing the article does not call for. Run it before reporting any task complete, and paste the output. If a check fails, fix the file, not the check.
In this repository4 other files it gives its agents.
- AGENTS.mdAGENTS.md
- templates/CLAUDE.mdCLAUDE.md
- intent-templateSkill
- secure-api-reviewSkill
The file
We split this CLAUDE.md into the rules it gives an agent
9 rules. About 0 are common, 7 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
Test: make test (runs scripts/validate.sh.
- Very specific: line 20
Every file has a layer: V verbatim, C completed, I install, R this repo.
- Distinctive: line 36
Do not add anything the article does not describe.
- Distinctive: line 38
If it is not in the article at all, it does not go in.
- Distinctive: line 39
Do not edit verbatim files to make them "work".
- Distinctive: line 42
Do not invent example projects, authors, or incidents.
- Distinctive: line 47
Test: make test (all checks pass.
- Distinctive: line 49
Run it before reporting any task complete, and paste the output.
- Distinctive: line 50
If a check fails, fix the file, not the check.
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
ai-native-sdlc-playbook
Companion to Anthropic's "The AI-Native SDLC playbook" (Louis Claxton, 2026-08-21). Three layers: the article's files verbatim (hash-locked), the pieces the article describes in prose implemented, and a plugin plus installer that place them into another repo. Nothing the article does not call for.
Commands
- Test: make test (runs scripts/validate.sh; ends with "validate: all checks passed")
- Lock: make lock (rewrites verbatim.lock; only after re-checking the article)
- Article drift: scripts/check-article.sh (network; not part of make test)
Conventions
- One change = one directory intent/
<NNNN>-<slug>/ holding intent.md, spec.md, plan.md, committed in that order, one commit each. The pre-publication record is on the history branch and tag v0.1.0; 0003 to 0008 landed together in the v0.2.0 commit. - Every file has a layer: V verbatim, C completed, I install, R this repo. docs/article-map.md ties each C file to the article sentence it implements.
- Verbatim files are listed in verbatim.lock and .claude/protected-paths.txt. They are never edited; make test fails if one changes.
- Templates live in templates/. Skills and commands reference them.
Architecture
- .claude/ is both this repo's live config and the plugin's component root; .claude-plugin/ holds the manifests. hooks.json wires all four hooks for the plugin; settings.json (verbatim) wires only the gate here.
- install/adopt.sh, driven by install/MANIFEST, is what /sdlc-init runs.
- scripts/ holds the validator, the hash check, the detector and the tests.
- AGENTS.md and .agents/skills are symlinks to CLAUDE.md and .claude/skills. Edit the targets, never the links.
Things Claude gets wrong
- Do not add anything the article does not describe. If it is in the article as prose, it is layer C and needs an intent/ triple and an article-map row. If it is not in the article at all, it does not go in.
- Do not edit verbatim files to make them "work"; the file map says what an adopter replaces. The production gate blocks any Bash command that contains both "deploy" and "production", including heredocs; use Write.
- Do not invent example projects, authors, or incidents.
- BSD sed and awk: no \s, no newlines in awk -v. CI runs on Linux, dev on macOS.
Verifying your work
- Test: make test (all checks pass; never skip or delete a failing check)
Run it before reporting any task complete, and paste the output. If a check fails, fix the file, not the check.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

