context-specs-factory-ai

Validates a mainspec and its slices via 3-subagent consensus plus expert review, then *applies* impactful fixes directly to the spec files. No human-in-the-loop summary or approval — agent-first. Touches `specs/<feature>/.validated` as its final committed action.

Reads a PRD (`prds/<feature>/prd.md`) plus its executable `run-prd-test.sh` (and any helper artifacts under `prds/<feature>/`), grounds them in codebase research, and produces `specs/<feature>/mainspec.md` plus dependency-ordered slices. Encodes the runner as a slice success criterion so implementation completion implies `./prds/<feature>/run-prd-test.sh` exits 0. Touches `specs/<feature>/.planning-done` as its final committed action. Agent-first — no human-in-the-loop.

Implements a single slice with Signal validation and unit tests. Agent-first — invoked by the slice-implementer subagent (under implement-mainspec). No human-in-the-loop; signal validation iterates up to a bounded `max_signal_iterations` (default 3) before reporting FAILURE.

Implements a mainspec end-to-end by auto-detecting mode. Sequential mode (≤3 slices) commits slices in order on the current `feature/<feature>` branch. Parallel mode (>3 slices) uses dependency-aware tiered execution with per-slice worktrees, branches, PRs, and auto-merge into the feature branch. Agent-first — invoked headless by the harness dispatcher with the feature slug as its single argument. No human-in-the-loop, no approval gates.

Meta skill for creating new SDD experts from domain documentation. Provide a markdown file path with domain knowledge + optional steering context. Automatically synthesizes references, generates signal scripts, validates, and registers in experts.md/signals.md catalogs. Single-prompt creation with follow-up tweaks. Triggers - create expert, new expert, make expert, expert for, domain expert, add expert (project)

One-time, guided setup of a standalone LLM-maintained wiki — a Karpathy "LLM Wiki" style knowledge base for a problem domain and your general architecture best practices. Scaffolds an external wiki vault (its own git repo) with /ingest, /query, /lint commands and a conventions doc. Use when a developer wants to start, create, bootstrap, or initialize a wiki / second-brain / knowledge base to understand a problem space before building. The front of the Human Loop's Understanding phase.

Turn an open-ended idea into a PRD plus a runnable definition of done. Use when a developer wants to start a new feature, "file a PRD", "spec out" or "scope" an idea, think through what to build, or kick off the harness for a piece of work. Produces prds/<feature>/prd.md + an executable prds/<feature>/run-prd-test.sh on a prd/<author-slug>/<feature> branch. The one human-attentive skill in the chain.

Evaluate the build trail of a PR — read the claude -p sessions the harness ran to build it, find where the project's context (Expert / AGENTS.md / skill / spec) served or failed the agents, then capture the learnings as evals (regression tests over the harness's own skills/context) and context fixes. Use when resolving a STUCK (diagnosis-first), auditing how a converged PR was built, or auditing a /learn memory PR. Human-driven and conversational — the trail-evaluating sibling of /evaluate-pr. Outcomes land on a branch (the PR's, or a fresh capture branch if you'll discard the PR) and reach memory via merge + /learn. Triggers - evaluate-sessions, evaluate sessions, review the build trail, diagnose stuck, audit how this was built, session observability.

Evaluate a PR the harness produced — walk the change, run the system together, and build a firm understanding before you merge it. Use after the harness hands a converged PR to you for review (the "Ready for your review" comment), or any time you want to deeply review an agent-authored PR. The human-attentive skill at the back of the chain; the mirror of /intent. Outcomes — merge, close, or fix-it-yourself-and-push - no handing work back to the loop.

Update the project's long-term memory after a merge to main. Reads the merged diff and the current memory, then writes Expert shards, discovered invariants, candidate lints, and AGENTS.md pointers — all on a reviewable learn/<sha> PR. Use post-merge (the harness invokes it automatically) or with --rebuild to regenerate memory from scratch. Triggers - learn, expert-update, update memory, update expert, post-merge memory, self-improve (project)

One-time, guided setup of the local agent-first coding harness — the polling loop, the deterministic dispatcher, its config, the AGENTS.md contract, and the worktree provisioning that makes per-feature worktrees runnable. Use when a developer wants to set up, bootstrap, install, or initialize the coding harness for the first time in a project.

Patch the code so `scripts/local-checks.sh` passes, after /implement-mainspec — a narrow post-implement polish specialist. Reads the (remediation-rich) check failures, fixes the underlying cause, never silences a check, re-verifies, and commits. Invoked headless by the dispatcher's two-strike local-checks gate; the dispatcher owns the retry counter. Triggers - fix-local-checks, fix lint, fix local checks, fix typecheck, make checks pass, post-implement polish (project)

Respond to an automated reviewer's findings on the current feature PR — a narrow, headless responder. Triages each reviewer finding into Clear / Ambiguous / Complex / Out-of-PRD-Scope, then acts — fixes the cause and pushes (diff-only) for Clear, posts one in-thread reply for the rest. Skips findings it already handled. Invoked headless by the dispatcher's feedback gate; the dispatcher owns the round counter and the STUCK escalation. Triggers - address-feedback, address review comments, respond to PR review, handle reviewer findings, PR feedback loop (project)