Unify repository and skill identity as Stash

[dd3ok]

Summary

• unify the public identity as Stash: npm @dd3ok/stash, repository dd3ok/stash, Claude marketplace dd3ok-stash, and display name Stash
• derive generated adapter versions from package.json, clean-regenerate adapters, and enforce lint-before-build in CI and prepack
• reject legacy three-part branding in source content and paths while excluding generated/runtime caches and Git worktree gitfiles
• replace partial compact-name substring promotion with whole-term sequence matching
• tune routing profile 3 for the known dense-description threshold cliff and add positive/negative golden coverage
• add distribution identity/version contract tests and regenerate all vendor adapters

Validation

• npm ci — 0 vulnerabilities
• npm run test:all — 39 core + 7 distribution tests passed
• npm run bench — 1,000 records; all exact/search/index budgets passed
• npm run pack:check — @dd3ok/stash@0.1.1 dry-run passed with the corrected prepack order
• skill validator passed
• root Codex plugin and generated Codex adapter validators passed
• real 176-skill catalog: ui false activations reduced from 25 to 10 relevant UI skills; go reduced from 6 false activations to no match
• path/content, ignored-cache, orphan-adapter, and worktree-style .git sentinels exercised directly

Pull request evidence

• Behavior changed: public identity, generated-artifact lifecycle, short-query phrase boundaries, and dense-description material routing
• Tests added: identity/version distribution contract plus positive, nearby-negative, negative-example, and no-match routing goldens
• Routing metric impact: repeated same-machine Windows runs were noisy but remained far inside the generous budgets; representative-catalog precision improved for the reported ui and go failures without changing all-relevant pagination semantics
• Platform impact: Windows NT 10.0.26200 with Node 24.18.0/npm 12.0.1 passed all local gates; GitHub Actions Ubuntu passed Node 20/22/24 and packaging/performance; macOS was not live-tested, and no platform-specific runtime path or catalog behavior changed
• Generated artifact drift: lint:artifacts passed before build; stale/orphan adapter sentinels were detected and clean regeneration restored the tree
• Vendor binaries: skill/plugin schema validators and generated contract tests passed; Claude Code and Antigravity CLIs were not installed for live invocation testing, and the Codex CLI could not be launched from the desktop package in this environment
• Security impact: no runtime network, telemetry, execution, or catalog mutation was added; the new brand scan is build-time/read-only, and adapter deletion is scoped to the repository's fully generated adapters/ tree

Cross-review

Reviewed iteratively with ChatGPT Pro until all P0/P1 disagreements were resolved. Final verdict: APPROVED — mergeable.

Non-blocking follow-up

A future routing profile can make dense-description evidence monotonic when an incidental group/tag term also matches; this pre-existing edge case is not exercised by the reported catalog failure and is not a merge blocker.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.