manoldonev
UserNo proof, no merge. Open-source Claude Code plugins for autonomous development: the agent asks before it builds, proves what it did, and cannot merge its own work. MIT, on the subscription you already pay for.
Categories
Indexed Skills (22)
audit
Read-only query of the current Claude Code session's tool-call ledger. Surfaces what tools Claude actually invoked (vs. what Claude claims it invoked). Observability only — a visibility signal for manual review.
design-faithful-spec
Normalize a Claude Design handoff into a faithful frontend spec for the repo — completeness inventory, behavioral/state contract, and design→real-stack component map. Use to turn a screen-level design handoff into an implementable FE spec before writing code. Dispatched by the intake router's design-handoff route, or by a session's choice (the design-sync engine that used to dispatch it was retired in
design-faithful
Implement a screen/component in the repo's FE app with high visual fidelity to a Claude Design handoff — mirror the nearest analog, reuse the repo's primitives and tokens, then live-render self-verify against the bundled screenshot and commit. Use to turn a design-faithful-spec (or a handoff) into committed FE code. Dispatched by a session's choice under the outcome-gated lean lane (the design-sync engine that used to dispatch it was retired in
pipeline-retro
Post-run retrospective for a dev-pipeline run: independent eval re-scoring, contract-deviation audit, and improvement routing. Run after a /dev-pipeline:run-lean run completes (or aborts); also reads the pre-#348 staged-run corpus.
pr-revision
Address human PR review comments — classify, auto-fix, clarify, or push back with evidence
run
DEPRECATED — kept as an ablation control and rollback lane; new work routes /dev-pipeline:run-lean. Fully autonomous pipeline: GitHub issue → branch → implement → review → PR
decomposition-reviewer
Reviews a set of already-scoped sub-issues for cross-ticket consistency, convention adherence, operational implications, and coverage gaps. Use after intake-orchestrator has produced tickets.
intake-interviewer
Interviews the user on an unstructured bug report or rough feature idea until the output would pass spec-reviewer (feature) or is reproducible (bug). Emits a GitHub-issue-ready body.
intake-orchestrator
Orchestrates spec review + scope decomposition for the dev-pipeline. Dispatches sub-agents, evaluates findings critically, decides whether to split work into parallel or sequential sub-issues.
intake
Manually triggered via /intake-toolkit:intake — the front door to the elicitation surface. Give it anything (a GitHub issue number, a pasted blob, a rough idea, an existing plan, a design handoff) and it classifies the scenario and routes to the right skill. Dispatch-only; contains no elicitation protocol of its own. Do not auto-select this skill; the routed skills carry their own triggers.
interviewing-baseline
Shared interviewing protocol for all intake-role skills (intake-interviewer, plan-interview, grill-me, intake-orchestrator escalations). Provides the interview loop rules and the Decision Ledger contract — schema, provenance enum, explicit-empty form.
plan-interview
Use BEFORE writing any implementation plan for behavior-changing work — plan mode, a direct feature/fix ask, or pre-flight for a dev-pipeline ticket. Surfaces the load-bearing design decisions to the engineer while the plan is forming and emits the Decision Ledger the plan gates require. Not for typo/mechanical fixes.
mutation-review
Propose→execute mutation-review protocol for co-located unit tests — mock-boundary discipline, assertion-strength anti-patterns, the propose/execute split, the (AC-n) test-title convention, and the blocker-mutant emission contract. Runner commands and repo-specific blocker mutants come from the consumer config and extension files.
review-lead
Orchestrates parallel code review across specialized reviewers. Use when reviewing code changes, PRs, or before committing.
reviewer-baseline
Shared review protocol for all code reviewer agents. Provides confidence scoring, finding classification, suppressed findings format, and standardized output structure.
local-dev-refresh
Refresh THIS MACHINE's second-shift plugin state - update the second-shift marketplace catalog, then every plugin installed from it ('plugin update' is the verb that actually upgrades - 'install' no-ops as "already installed"), realign the current repo's project-scope stragglers - reporting rather than refreshing the ones a user-scope record already makes redundant - print the before/after version delta, remind to restart. Scoped to the second-shift marketplace only - other marketplaces on the machine are left untouched. Run after merging plugin bumps (dogfooding) or after a release/pin change.
onboard
Onboard the current repo onto the second-shift marketplace - detects tracker/topology/commands with provenance, drafts the config for one accept-or-edit review, writes settings pin + config + lockfile, validates with config-lint. Run from the target consumer repo. Requires jq, gh (authenticated), git.
perf-retro
Cross-run execution-latency retrospective for dev-pipeline runs: fidelity-triaged timing profile, ranked optimization candidates with named regression guards, and improvement routing. Run periodically over the recorded run corpus.
review-lean
The REVIEW half of the lean lane — review an open lean PR from a fresh top-level session and produce the committed verdict record. Runs outside the build session by design; the build harness cannot produce this record.
writing-tests
Use when adding or changing a test in this repo — the tier map (where a new guard goes), the scenario-first rule, the no-prose-presence-guards and no-mirror-harnesses rules, the mjs-seam grep exception, and the mutation-catalog anchoring obligations.
grill-me
Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".
doctor
Verify this repo's second-shift install/config state against the committed lockfile - never-installed, enabled-but-not-installed, version drift (behind AND ahead), project-scope records a user-scope one makes redundant, ref-less marketplace shadowing, skill/agent shadow collisions, opt-outs, config-lint. Prints exact remediation commands, scoped to the record that actually loads. Run after cloning, after upgrades, whenever the toolkit feels absent.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.