Anchor Point

A model-agnostic documentation operating system for solo devs running multiple side projects.

Current version: v3. See CHANGELOG.md for what changed.

Drop this folder into an AI project, repo, or knowledge base, then point tool-specific adapters at it. The agent becomes the doc specialist. Knows where everything goes. Keeps your docs from getting messy. Battle-tested, distilled from cross-project mining of real-world doc drift incidents.

---

Who this is for

You bounce between a handful of projects. Every agent session starts with you re-explaining what you're working on for ten minutes. Your docs are stale. They look different in every project. And the model keeps re-discovering the same stuff you already wrote down somewhere.

Sound familiar? You don't need another tool. You need a system.

What's in here

• A clean way to organize your docs. 5 main files at the root, plus a docs/ folder for everything else.
• Naming rules that work the same way in every project (so you stop having "is it auth.md or Auth.md or auth_flow.md?" arguments with yourself).
• A checklist of the common ways docs go sideways, so you catch them early.
• Templates for each of the 5 main files. Drop-in, fill-in.
• Six workflows for the project lifecycle: Init (new project), Review (start of session), Update (end of session), Audit (periodic cleanup), Project Ratchet (duplicate-only improvement loop for one project's docs), and Doc Workflow Ratchet (hardening the doc workflow itself from project evidence).

Quick start (under 5 minutes)

1. Drop this folder into an AI project or repo knowledge folder (Claude Project, Codex skill folder, Cursor rules folder, or a plain docs repo).
2. Start a chat in that project.
3. Try one of these to see it work:
   • "Init docs for a new Next.js project called my-app"
   • "My project's AGENTS.md is at /path. Review it and tell me what's drifting."
   • "Just finished a coding session. Help me write the handoff."
   • "My docs are a mess. What's the cleanup plan?"

The agent will respond as a doc specialist and walk you through the right workflow. That's it.

Versioning

Anchor Point uses whole-number public versions. The current published shape is v3. The next accepted edit is v4, then v5, and so on. The changelog explains whether a change was tiny, naming-only, structural, or a major rethink; the version number itself does not encode patch/minor/major semantics.

Older decimal labels (v3.4.1, v3.4.0, etc.) are retained in the changelog only as historical audit detail.

What's in this folder

File	What it's for
SKILL.md	The operational skill. Modes, workflows, hard rules, drift checks, scoring rubric, naming, where stuff goes.
identity.md	Who the specialist is. Their background, what they're good at, what they won't touch.
examples.md	A few example interactions. Shows you what good output looks like.
components/	Internal building blocks (snapshot, verify, harvest, promote-memory, security-preflight, workflow-score).
reference/	The full spec, naming rules, drift checklist, where-does-this-file-go decision tree, ratchet/autoresearch specs, and templates.
README.md	This file.

Vocabulary

Use these names going forward:

Term	Meaning
Capability	A repoed, model-agnostic package of docs, rules, routines, and references. Anchor Point is a capability.
Workflow	A user-facing process such as Init, Review, Update, Audit, Project Ratchet, or Doc Workflow Ratchet.
Routine	A chainable repeatable block inside a workflow, such as harvest, verify, snapshot, score, or security preflight.
Component	A reusable internal checklist, rubric, template, or scoring rule.
Adapter	A tool-specific wrapper, slash command, installed skill, prompt, or scheduler entry that points back to the capability.
Overlay	Project-specific context that stays local to the project instead of becoming a global rule.

Tool-Specific Adapters

AGENTS.md is the canonical AI bootstrap file. If a platform needs a different entrypoint, create a thin adapter that points back to AGENTS.md and this capability rather than forking the rules.

Examples: Claude Project knowledge, a CLAUDE.md shim, Codex installed skill metadata, Cursor rules, slash commands, or scheduled automations.

The whole thing in one sentence

Every project gets the same 5 root docs (README.md, AGENTS.md, STATUS.md, ROADMAP.md, LOOKUP.md), plus a docs/ folder with 9 known subfolders. Each file does one job. Anchor Point keeps you from drifting from that.

What your project looks like after Init

your-project/
├── README.md              ← the public face
├── AGENTS.md              ← rules + folder map + current phase (AI reads first every session)
├── STATUS.md              ← what just shipped + what's next (pure-function handoff)
├── ROADMAP.md             ← priorities + next decisions
├── LOOKUP.md              ← topic-driven lookup ("where do I find X?")
└── docs/
    ├── DOCS-INDEX.md      ← location-driven file map (created when docs/ has >10 files)
    ├── status-history/    ← rotated STATUS.md backups (dated)
    ├── roadmap-history/   ← rotated ROADMAP.md backups (dated)
    ├── decisions/         ← one file per decision (dated)
    ├── reference/         ← verified-state files (config truths, capability matrices)
    ├── playbooks/         ← process runbooks ("when X happens, do Y")
    ├── dev/               ← architecture, feature design, schema docs
    ├── release/           ← release notes, migration steps
    ├── reviews/           ← code review reports, security audits
    ├── research/          ← experiments, exploration drops
    ├── _archive/          ← historical (dated batches with READMEs); appears on demand
    └── _private/          ← gitignored sensitive content; appears on demand

All 9 standard docs/ subfolders are created at bootstrap, each with a README.md stub that documents what goes there. Empty folders carry signal, not noise. The _archive/ and _private/ administrative folders appear when their content arrives.

The 6 workflows (lifecycle)

NEW PROJECT        →  Init        bootstrap all files
START OF SESSION   →  Review      read state, suggest next steps (read-only)
END OF SESSION     →  Update      handoff, roadmap sync, harvest, verify, snapshot
PERIODIC CLEANUP   →  Audit       deep alignment, file relocation, drift score
LOW SCORE / MESSY  →  Project Ratchet      duplicate-only cleanup of one project's docs
SYSTEM HARDENING   →  Doc Workflow Ratchet improve the doc workflow itself from evidence

You don't memorize commands. Just talk naturally ("catch me up", "wrap this session", "my docs are messy") and the right workflow kicks in.

Why the name

Docs drift. Across sessions, across projects, across AI tools. Same conventions get re-explained. Same gotchas get re-discovered. Same questions get re-asked.

An anchor point is the fixed thing everything else ties to. That's what this folder is, but for your docs.

License

MIT. Take it, fork it, change it, ship your own.

Built on the Interpretable Context Methodology

This follows the Clief Notes "folders as architecture" idea. Each file does one job. The structure tells you what's where. Someone with no context can clone it and use it cold.