Engineering skills: SQL, clean architecture, clean code, DDD, database optimization, chaos engineering, Spring Boot, Kotlin, TDD, testing, bug diagnosis, security audit, architecture review, deprecation & migration, SRE, performance profiling & optimization, incident response, documentation strategy, like-my-code (a repo's style as harness conventions, each rule with evidence and a cited source). ADR drafting moved to write:plans (ADR format); diagram moved to write:diagram.
36 skills · hooks
A sibling of graph for requests too big for one pass: a TaskManager MCP (teams) sizes the request; size L is planned into a PRD and user stories, split into packages that each run the full harness in their own worktree, integrated, QA'd and gated by a background daemon; size S is handed to the development harness. Entry skills teams:orchestrate/develop/document/plan/qa/sprint, board and ticket views (tm_board, teams:board). Runs are kept under .teams_output/ and ~/.harness/tasks/. Zero runtime dependencies beyond Node 18+.
15 skills · 1 MCP server · hooks
Portfolio and career skills: manage (the portfolio deck as one workflow — material, skeleton, feedback, deck, versions), material (collect handed material), fit (JD or company type), portfolio feedback, rewriting and JD tailoring, pattern analysis, mock interview, interview plan, job-application-workflow (through a 자기소개서 step via write:writer-verification), deck-builder (reference template.pptx + deck.mdx → pptx; recolors a pptx to a brand palette), resume-builder (resume-deck.mdx → the user's own .docx, PDF, page fit, final AI-trace check), and render-review (page-by-page review of a rendered deck or resume).
12 skills · hooks
Thinking quality skills: assumption extraction, bias auditing (v2 — context probe, per-bias remedies, calibration check), fallacy detection, mental models, epistemic reasoning, tradeoff articulation, question upgrading, and second-order thinking.
10 skills · hooks
Writing skills: plans (gate-ready implementation plans and ticket breakdowns with production-time gap/ambiguity checks + per-step pass bars; technical blog posts; SBI feedback), prd (PRDs, design docs, RFCs and proposals co-written section by section with reader testing), adr (ADRs and fixed-template design reviews), requirements (a requirement or 기획서 split into actors, rules, open questions with defaults, and items with Given/When/Then criteria),spec (a settled conversation synthesised into a spec, open points marked, no re-interview), skill writing, and writer verification (v2 — five passes incl. a humanizer for machine tells, draft loop that rewrites until 🔴🟡 = 0, PR descriptions from a diff as first-class input), and tech-book (performance-based book arc, cumulative examples, transfer practice, six required teaching elements within a chapter length budget, staged review loops with defined severities, lab-running fact checks, pinned chapter end states and a cap closure pass, and PDF in bundled NanumGothic), diagram (checked self-contained SVG + PNG figures: page layout, B-tree, lock timeline, ER, chart, architecture, sequence; moved from develop). Knowledge-base, graph, RAG, and query skills moved to the knowledge plugin.
10 skills · hooks
Answerability-gated knowledge skills and local MCP: linked Markdown, relation-aware graph extraction, graph visualization, RAG preparation, SQLite index building, and source-grounded retrieval.
8 skills · 1 MCP server · hooks
Thinking and ideation skills: mentor (any-field mentoring — takes a position, tests yours, hands over the judgment rule), brainstorming (≥3 options before judgment, kill-criteria convergence, hands off to write:plans), grill (a held plan interrogated: repo read first, one numbered round with recommended answers), devils-advocate (steel-man first, core vulnerability + reversibility), redefine-problem, back-to-basics, untangle-thoughts, and negotiate (counsel on your side).
8 skills · hooks
Agent orchestration skills: dispatching parallel agents as a persona allocator (fan out independent jobs, match each to a best-fit specialist), subagent-driven development, task decomposition, and the other model as a read-only second opinion or, on request, a worker with permissions bypassed (use-codex, use-claude).
5 skills · hooks
Harness — a lightweight reasoning floor. Pass a raw request; a fixed six-stage engine runs Plan(opus) → SetGoal(opus, adversarial critic) → Implement(provider-routed, Codex when enabled) → Test(provider-routed, Codex when enabled) → QualityGate(opus, bounded loop + goal-level gate) → Report(sonnet). Filters repetition and below-threshold answers regardless of main-session model. B (request + engine) default; A (bespoke Workflow) for custom control flow. Enforcement: opt-in PreToolUse gate — projects list gated paths in .claude/harness-gate.json; edits there require harness engagement (fail-open). The install skill scaffolds a project: it embeds the self-contained gate hook (copies goal-gate.mjs into .claude/hooks/ and merges a PreToolUse entry into committed .claude/settings.json), plus gate config, default conventions (coding/verification/boundaries), and an ambient CLAUDE.md section — project-owned copies, never overwritten. It also optionally embeds the engine + Codex CLI adapter/runner + statically-referenced skills into .claude/harness/ for plugin-less environments (opt-in; dynamic subgoal skills[] cannot be pre-enumerated). Mode M: for control flow the fixed stages cannot express, the harness generates a bespoke Workflow from templates/meta-skeleton.js (contract preserved: judge≠actor, provider routing, bounded loops, deterministic Test, goal-level gate). Workflow path: when codex_provider is auto or required, every Implement/Test stage uses a minimal Sonnet controller to resolve engine/codex-exec-adapter.mjs and delegate to a separate local codex exec --json process; implement_provider/test_provider codex are optional trace hints, not routing prerequisites. On Codex success the controller only converts Codex output into the normal HANDOFF/evidence shape; it does not redo implementation or verification. codex_provider=auto allows explicit degraded Sonnet fallback; codex_provider=required reports provider failure instead of silently falling back. DW-off/Workflo
5 skills · hooks
Skill management: create skills in the repo's house identity, quality-assure them (6-dimension QA), audit trigger coverage, audit a workspace for what it already has before installing, and retro (a session's repeated mistakes turned into checks).
5 skills · hooks
Read-only database access: the database-readonly MCP server (MySQL; session READ ONLY on every connection, statement whitelist, row cap) and skills login (macOS password dialog into Keychain, 1Password op:// optional), logout, profile (local/dev/stage/prod; no confirmation, read-only) and query (personal-data columns hidden).
4 skills · 1 MCP server
Graph engineering for the harness flow: a local MCP server owns graph state, routes nodes to vendor CLIs or the session, adjudicates results against the worktree, and keeps payloads on disk. Includes graph:install and graph:orchestrate.
2 skills · 1 MCP server · hooks
Verification before completion: stand up a verify-level (pass bar), then confirm it with an isolated subagent so the check is independent of the context that did the work.
1 skill · hooks
Planning skills: executing-plans (pre-flight plan gate — critically review a plan, set a QualityGate pass-bar per step, then route to the right executor).
1 skill · hooks
What a session left behind and a safety net: a /session pane with files changed, commits, denied calls and the longest gap, a retro band on the next start, a safe stop for stray claude -p children, guards that confirm destructive commands and secret-file writes, /memo notes pinned to the model's context, /smart-compact (recap at a set context %, then compact), /handoff, /recap and /lessons from that recap, a /task timer and the session cost in the status line, a /board of every open session with its harness, graph and teams runs, and a message to the other sessions of a repo when main moves.
hooks
Steam-style achievements for using this marketplace's skills: a toast on unlock, an /achievements pane, trophy:list in conversation, and a view of which skills your prompts should have triggered, and a local record of skill failures. Opt-in anonymous daily counts and error codes (off by default).
1 skill