done-or-not

Done or Not

Check whether "done" is actually done.

Done or Not is a Claude Code plugin for auditing whether claimed AI-assisted feature work is actually implemented.

It is markdown-first and audit-only: it reads a repository, compares claims against implementation evidence, and can save audit memory under docs/implementation-audit/ in the target repo.

Current version: v0.1.2

What It Does

• discovers project semantics before judging implementation
• builds a candidate feature map
• separates feature presence from implementation authenticity
• audits one confirmed feature through an evidence chain
• checks merge readiness or main-baseline readiness
• proposes or saves audit memory under docs/implementation-audit/ with approval

What It Does Not Do

• does not modify product code
• does not add tests
• does not fix bugs
• does not change schemas
• does not change CI
• does not commit, merge, or deploy
• does not act as a global sandbox

Requirements

• Claude Code CLI

Installation

From the CLI:

claude plugin marketplace add FoFxjc/done-or-not
claude plugin install done-or-not@fofxjc

From within a Claude Code session:

/plugin marketplace add FoFxjc/done-or-not
/plugin install done-or-not@fofxjc

Updating

From the CLI:

claude plugin marketplace update fofxjc
claude plugin update done-or-not@fofxjc

From within a Claude Code session:

/plugin marketplace update fofxjc
/plugin update done-or-not@fofxjc

Usage

Primary commands:

/done-or-not:discover-project-semantics
/done-or-not:propose-audit-docs
/done-or-not:save-audit-docs
/done-or-not:audit-feature
/done-or-not:audit-merge-readiness

Typical flow:

Discover project semantics
  -> propose audit docs
  -> save audit baseline
  -> audit one feature
  -> audit merge readiness when needed

How It Works

Done or Not is a lightweight single-repo completion legitimacy layer for AI-assisted feature work.

It validates whether claimed feature and project implementations inside a repository are actually real.

It does not call a feature fake just because a code path looks suspicious. It first establishes project semantics, then allows stronger authenticity judgments only after scope is confirmed or strongly evidenced.

Core rule:

No semantics -> no fake verdict.
Draft semantics -> preliminary audit only.
Confirmed semantics -> real authenticity gate.

Generated Audit Folder

docs/implementation-audit/
├── README.md
├── feature-map.md
├── implementation-semantics.md
├── placeholder-policy.md
├── git-context.md
├── audits/
├── decisions/
└── merges/

This folder is the project audit trail. It does not replace official product docs.

Feature Status Model

Done or Not uses a split status model:

Model Part	Allowed Values
Presence Status	Claimed, Observed, Inferred, Confirmed
Authenticity Status	Not audited, Preliminary, Partial, Mostly real, Real, Non-operational, Fake, Drifted

Discovery and proposal phases never use final verdict language. During discovery, features keep Authenticity Status: Not audited.

Local Development

Validate the plugin manifest:

claude plugin validate .claude-plugin/plugin.json

Validate the marketplace manifest:

claude plugin validate .claude-plugin/marketplace.json

Load the plugin directly for local dogfooding:

claude --plugin-dir /path/to/done-or-not

If /plugin opens the generic plugin management UI and does not clearly list commands in your terminal, inspect the commands/ directory or run a known namespaced command directly.

If claude --plugin-dir fails with an EPERM error involving ~/.claude.json, that is a local Claude CLI config or permission issue, not a plugin validation failure. Some sandboxed runners also cannot write under ~/.claude*, which can block local plugin loading even when the plugin itself is valid.

File Structure

done-or-not/
├── .claude-plugin/
│   ├── plugin.json
│   └── marketplace.json
├── commands/
├── skills/
├── templates/
├── docs/
│   ├── positioning.md
│   └── roadmap.md
├── CLAUDE.md
├── README.md
└── LICENSE

Positioning

Done or Not is a lightweight single-repo completion legitimacy layer for AI-assisted feature work.

It validates whether claimed feature and project implementations inside a repository are actually real.

It is not an Agent OS, orchestration engine, workflow platform, memory system, task tracker, or multi-agent controller.

It focuses on repo-local feature claims, implementation evidence, authenticity verdicts, and audit records.

Roadmap

See docs/roadmap.md for the path from the current plugin to v1.0. The roadmap is the source of truth for milestone scope and non-goals.

License

Apache-2.0