Agent-Skills

Analyzes bounded skill evidence, classifies root causes, and recommends a lifecycle lane such as keep, observe, improve through Skill Factory hardening, merge with approval, or retire with approval. Use when a skill is not working, a skill is not triggering correctly, evals or Tessl disagree, repeated failures need debugging, or skill performance issues need evidence-backed repair handoff items.

Scan Codex session history for skill failures, usage patterns, and coverage gaps. Use when the user wants daily skill-health monitoring or evidence-backed recommendations about installing, improving, merging, or pruning skills.

Scan Codex session history for skill failures, usage patterns, and coverage gaps. Use when the user wants daily skill-health monitoring or evidence-backed recommendations about installing, improving, merging, or pruning skills.

Fixture skill with promotable eval scenario metadata.

Minimal valid Codex skill source fixture for JSC-391 scaffold tests.

Analyze options and trade-offs before traceable work is ready, without pretending exploration is a plan. Use when the user wants options, alternatives, decision matrices, or a bounded ideation pass before trace planning.

Generate closure proof with exact commands, outcomes, artifact paths, untested lanes, and residual risk. Use when work and review need a receipt before closeout, release, tracker movement, or handoff.

Plan how to implement one approved slice, including file targets, validation gates, rollback, and proof lanes. Use when a slice spec is accepted and the user needs the implementation order before code changes.

Analyze stale or conflicting lifecycle state by separating local, PR, CI, review, tracker, artifact, and session truth. Use when status is confused, work resumed after drift, evidence lanes disagree, or a closeout claim may be stale.

Plan a stuck, stale, or unsafe frame into a smaller viable direction with rollback and decision boundaries. Use when a migration, plan, PR, or implementation path is producing churn, conflicting claims, or unsafe scope.

Create a durable guardrail, validator, doc, eval, or skill-reference update from a verified failure or learning. Use when the user asks to capture learning, prevent recurrence, encode review feedback, or refresh stale guidance.

Review the completed slice for correctness, risk, evidence gaps, and next-stage blockers. Use when the user asks for review, PR readiness, risk assessment, blocker identification, or post-work verification.

Create one selected trace or tracker slice spec with acceptance criteria, constraints, validation, and exit conditions. Use when exactly one slice has been selected and needs a buildable contract before implementation planning.

Plan traceable work bullets that connect intent, evidence, acceptance, and proof gaps. Use when an idea or reframe needs decomposition into traceable work units before tracker or slice specification.

Plan tracker-ready slices with dependencies, labels, acceptance fields, and validation placeholders. Use when trace work needs Linear-ready or tracker-ready issue structure without actually mutating the tracker.

Build only the approved slice while preserving scope, unrelated worktree changes, and validation evidence. Use when an approved execution plan or explicit bounded task is ready for implementation.

Use when the user asks about Shachar Azriel's AI Native DevCon talk on executable specs, verification layers for agentic coding, planner/verifier separation, requirement mapping, product-gap detection, and safe review architecture.

Use when the user asks about Macey Baker and Baruch Sadogursky's workshop-style AI Native DevCon session on turning repeated agent work into skills, rules, scripts, hooks, and evals.

Use when the user asks about Christopher Batey's talk 'Building Product Teams in the Age of AI: What We Had to Relearn Every Quarter' (Latent Space, 2026) — including questions about running AI-assisted product engineering teams, his three pillars (path to production at AI speed, training/evaluating AI-enabled engineers, designing workflow for parallel change), ADR-first workflows with agents, why review becomes the bottleneck, the producer 'black box' (harness/host/model), vanity metrics vs adoption, two-to-four-person sub-streams, one-complex-task-at-a-time, 'you build it, you run it, you drive adoption', or applying his approach to current work.

Answers questions about, retrieves safe excerpts from, explains concepts from, and summarizes key arguments in Birgitta Böckeler's talk "State of Play: AI Coding Assistants" (AI Native Dev conference, 2026). Use when the user asks about the last 12 months in AI coding assistants, model-task fit, LLM statelessness, context window and attention trade-offs, coding harnesses, harness engineering/context engineering, skills/MCP/sub-agents/plugins/hooks, guide-and-sensor feedback loops, background agents and swarms, review bottlenecks, AI coding costs, cognitive surrender, or risk-based supervision of coding agents.