Skills
Plain Agent Skills bring the same scope-and-proof habits to compatible coding agents.
portable promptsWorktreeProof is open-source vibe coding guardrails for AI coding agents: a small, local-first toolkit that keeps work scoped, changes inspectable, and every finished lane backed by a clear receipt.
Apache-2.0 · Node.js 20+ · zero runtime dependencies
Agents move quickly when the boundaries are explicit. WorktreeProof gives a lane a beginning, a safe working surface, and a verifiable end.
Turn a request into one named scope, target check, and owner-visible intent.
Keep parallel work apart with conflict-aware reservations and disposable worktrees.
Capture commands, test output, UI review notes, and the context needed to repeat them.
Ship a redacted record that says what changed, what passed, and what remains.
Use the pieces you need, from a single local command to a coordinated set of agent lanes.
Plain Agent Skills bring the same scope-and-proof habits to compatible coding agents.
portable promptsGive each lane a unique identity and non-overlapping file scope before it starts.
explicit ownershipUse clean, temporary working surfaces so parallel edits do not quietly collide.
isolated by defaultChoose the smallest local or connected capability that can complete the next step.
capability-awareKeep visual checks in the receipt, alongside keyboard and responsive evidence.
screen to proofUse focused checks and one clear terminal disposition instead of ceremonial status.
quality with intentKeep context bounded, reuse receipts, and avoid retries that add noise without progress.
less wasteLeave a compact, redacted record that another person or agent can inspect later.
closure you can readExchange bounded local task and result messages without launching either assistant or sharing hidden context.
explicit handoffSee redacted host-reported task status and mode while raw IDs, titles, prompts, and paths stay out.
reported, never inferredThese are patterns, not a new ceremony. Pick a recipe, adapt the scope, and let the receipt carry the context forward.
Reserve the failing area, reproduce once, and attach the focused test that proves the fix.
worktree-proof plan bug-fix --scope src/See the command path Pair a visual pass with keyboard and viewport notes before the lane closes.
worktree-proof run ui -- npm testSee the command path Collect checks and redacted evidence, then make the terminal disposition explicit.
worktree-proof close release --receipt proof.jsonSee the command path WorktreeProof is designed to be inspected before it is adopted. Install the tagged GitHub release, run the local doctor, and keep every write explicit.
Read the install notes$ npm install --global github:Nedal7707/worktree-proof#v0.4.0
$ worktree-proof doctor
$ worktree-proof resources scan --json
✓ local diagnostics ready — inspect before writingWorktreeProof keeps its state in the project you run it in. The toolkit has no runtime dependencies and no built-in telemetry by default; inspect the source and choose your own integrations.
Use ordinary care with command output and receipts: keep private values out of both.
A few straight answers before you open the repo.
WorktreeProof is an open-source toolkit for scoped AI coding-agent work. It helps teams reserve non-overlapping lanes, run checks, and record a closure receipt.
No. It makes the scope, checks, and evidence easier to inspect; your normal review and release decisions still matter.
The current project is a JavaScript and Node.js toolkit that can be run from a local checkout. It has no runtime dependencies.
The core workflow is local and model-agnostic. The repository also includes plain Agent Skills and adapters that you can inspect and adapt.
Read the docs, try a recipe, and send a focused change. Small improvements to schemas, skills, adapters, and examples are welcome.
Read contributing notes