working-skill-repo

Inspect live routes and manage optional user-local OpenAI-compatible/LiteLLM extra routes through the kbrouter CLI.

Single user-facing KB completion command. Takes a feature, plan, or manifest from its current state through planning, work, post-work finalization, and configured delivery: local completion, pushed PR, or explicitly configured direct integration and sync.

Configure optional portable per-project lower-tier-attempt and delivery policy.

Large-initiative coordinator for KB workflows. Use for app migrations, framework rewrites, major architecture changes, multi-subsystem initiatives, multiple brainstorms/manifests, long backlogs, or queued execution across related workstreams.

Internal post-work quality and learning phase. Runs kb-review -> resolution gate -> follow-up resolution -> proof/demo evidence -> compound -> learn -> evolve -> memory refresh/compact -> cleanup after kb-work finishes all slices. Normally invoked by kb-work or kb-complete; not the primary user-facing completion command.

Deprecated compatibility alias for kb-complete. Use when an older prompt or workflow says 'kb finish'; delegate the same input to kb-complete and report the kb-complete result.

Functional-test strategy and test-quality audit for KB workflows. Use when deciding whether a slice needs functional/e2e/browser/API workflow tests, when existing tests may be mocked theater, or when user-visible behavior must be verified without manual QA.

Durable objective governor for KB workflows. Use when the user sets a goal, wants work to run for days across sessions, says continue until done, asks for vDone, or needs a long-lived objective forced through KB routing, planning, work, finalization, delivery, and proof gates. Owns goal state and stop conditions; delegates execution to kb-start, kb-task, kb-epic, kb-work, kb-finalize, and kb-complete.

Compact the current session into a KB handoff restart packet for a fresh agent. Use when the user asks for a handoff, restart packet, fresh-session note, session transfer, pause/resume note, or wants the next session to continue from local project memory.

Internal KB integration phase. Applies an explicit direct or merge delivery policy after kb-finalize, verifies the remote default branch contains the delivered commit, and runs configured post-integration synchronization. Normally invoked by kb-complete.

Project-memory setup, lookup, and refresh skill for KB workflows. Use when starting in a repo, when another KB skill needs the right architecture docs without broad repo crawling, or when the user says "map", "project context", "where is this", "what should I read", "setup memory", or "bootstrap context". If memory is missing or badly stale, invokes kb-map-bootstrap.

High-cost KB memory maintenance pass for repo-local project memory. Use when docs/context/memory-maintenance.md says review is due, after large epics, after repeated contradictions/stale-doc signals, or when the user asks to deep review, consolidate, compact, or reconcile KB memory. Starts from recorded signals and pointers instead of blind repo crawling.

Break a brainstorm or feature into vertical-slice task plans with dependency DAG, verification strategy, and HITL flags. Default planning workflow for end-to-end vertical slices instead of horizontal phases. Use when the user says 'kb plan', 'plan', 'create a plan', 'plan this', 'slice this', 'break into vertical slices', or wants independently-grabbable tasks.

Internal checked-in PR delivery for reviewed KB work. Validates complete-to-ship, audits final scope, commits intentional files, pushes a non-default topic branch, and creates or updates a correctly based PR without merging. Normally invoked by kb-complete.

Default KB start/router. Use when the user says "kb", gives an idea or ambiguous work request, starts a fresh session, asks what to do next, or wants the workflow to choose the right lane without making the user pick ceremony. Delegates project-memory setup and lookup to kb-map before choosing a lane.

First-principles autonomous KB task runner. Use when the user says /kb-task, asks the app/agent to do a task from first principles, wants the agent to figure out the route and continue until done, or wants one bounded app/workflow change handled end-to-end without choosing between kb-fix, kb-brainstorm, kb-plan, kb-work, klfg, or kb-complete.

Autonomous troubleshooting and self-correction loop. Use when the user says troubleshoot, debug this, figure out what's broken, self-correct, run the logs, use Playwright/browser, iterate until fixed, or asks the agent to diagnose failing app behavior without hand-holding.

Bounded swarm executor for kb-plan output. Runs all ready vertical slices from the dependency DAG, parallelizing only independent/isolated work, with TDD enforcement, scope gates, and HITL pauses. Use when the user says 'kb work', 'work the plan', 'execute the plan', 'run the KB pipeline', 'execute all slices', or wants guided execution of a planned feature.

Deprecated compatibility alias for kb-complete. Existing 'klfg' and full-pipeline requests delegate to the single state-aware kb-complete command.

Capture and replay deterministic regression snapshots between KB slices. Use after a slice passes QA to freeze passing state, and before future slices start to prove earlier slice behavior has not regressed.

Shared phase-gate policy for KB workflows. Use before moving from brainstorm to plan, plan to work, work to complete, or complete to ship when P0/P1/P2/P3/P4 findings, review issues, ambiguity, weak tests, or unresolved risks exist.

Proportional brainstorming for vertical-slice work. Orients from repo memory before asking product questions, runs external research only when it is likely to change framing, and produces requirements for `/kb-plan`. Use when the user says 'kb brainstorm', 'brainstorm', 'brainstorm with research', or 'brainstorm before kb-plan'.

Promote mature instincts (confidence > 0.8) into full Copilot skills that get auto-discovered. Clusters related instincts and generates SKILL.md files in .github/skills/.

Deterministic verification harness for KB workflows. Use when code should be tested, linted, typechecked, built, security-checked, or validated by scripts instead of relying on LLM judgment; also use before kb-complete, kb-ship, or after kb-work slices.

Surgical fix loop for QA and lint failures. Progress-based retry with stuck detection and a 5-iteration ceiling. Called by kb-qa when checks fail — not invoked directly by users.

Extract reusable patterns from recent work into instincts. Run after completing features, fixing bugs, or at session end to capture what the project learned.

Quality assurance gate for all slices. Runs lint checks on every slice, browser verification on frontend slices. On any failure, invokes kb-repair for surgical fixes. Hard gate — the browser reports what rendered, the linter reports what's dirty, the model does not self-report.

Small bug-fix lane with bounded diagnosis, agent-owned verification, and escalation. Use when the user reports a bug, failing test, broken behavior, or asks for a narrow fix that should not require a full brainstorm or plan unless it grows.

Token-expensive bootstrap skill that deeply indexes a new or existing project and creates the standard KB memory layout. Use when kb-map reports missing or badly stale memory, when entering an existing project without todo.md/docs/context/PROJECT.md, or when the user says "bootstrap this project", "deep map this repo", "build project memory", or "index this app".

Claims-vs-code evidence review for repos, skills, eval harnesses, README claims, commit messages, and tests. Use when the user asks to audit claims, verify whether implementation matches documentation, find overclaims, review evaluator honesty, compare README/skills against code, or run a proof-focused critic pass before shipping.

Refresh stale or drifting learnings and pattern docs in docs/solutions/ by reviewing, updating, consolidating, replacing, or deleting them against the current codebase. Use after refactors, migrations, dependency upgrades, or when a retrieved learning feels outdated or wrong. Also use when reviewing docs/solutions/ for accuracy, when a recently solved problem contradicts an existing learning, when pattern docs no longer reflect current code, or when multiple docs seem to cover the same topic and might benefit from consolidation.

Document a recently solved problem to compound your team's knowledge

Lazy architecture exploration lane. Use when the user asks where a codebase should get deeper, simpler to test, or more modular, not for ordinary cleanup or diff review.

Honest, evidence-based dialogue with principled pushback. Activates whenever the user challenges, rejects, corrects, disputes, or pushes back on a recommendation, factual claim, plan, or judgment - even without explicitly requesting first-principles reasoning. Also triggers on "first principles", "don't just agree with me", "push back", "be honest", "challenge this", "devil's advocate", or when the user expresses frustration with sycophantic responses.

Structured KB review using tiered persona agents, confidence-gated findings, thermonuclear structural-quality review, and a merge/dedup pipeline. Use when reviewing KB workflow code changes before completion, before creating a PR, or when kb-complete needs its review gate.

Test-driven development with red-green-refactor loop using vertical slices. Use when building features test-first, fixing bugs with regression tests, or when the user says 'TDD', 'test first', 'red-green-refactor', or wants integration-style tests that verify behavior through public interfaces.

Use when creating durable work items, managing todo lifecycle, or tracking findings across sessions in the file-based todo system

Use when reviewing pending todos for approval, prioritizing code review findings, or interactively categorizing work items

Repo-native eval setup mapper. Use during kb-map-bootstrap, when a repo has no meaningful eval harness, or when the user asks how to evaluate an app/workflow. Detects app pattern, chooses the right proof tools, scaffolds one safe smoke eval when possible, and wires the command into KB testing docs/checks.

Structured code review using tiered persona agents, confidence-gated findings, and a merge/dedup pipeline. Use when reviewing code changes before creating a PR.