sam-skills

Delegate a bounded implementation task to the fixed Grok worker (grok-4.5) under workspace sandbox and headless execution. Use when a host agent or user needs Grok to implement, fix, or verify a scoped coding task; default effort high unless explicitly supplied.

Create and validate risk-based Playwright browser tests for changed or reported user flows, including real linked UI/backend proof, exact route and network assertions, permissions, persistence, and optional local video evidence. Use when asked for Playwright, browser E2E, UI regression tests, cross-browser flows, route or CORS verification, or explicitly requested PR/MR test evidence.

Record and validate a human-paced MP4 walkthrough of a completed task, feature, or bug fix using the real runnable UI, with acceptance-criterion traceability, privacy checks, deterministic media validation, and optional external publication. Use when asked for a demo video, proof video, walkthrough, screen recording, MP4, or explicitly requested PR/MR video attachment.

Consult Codex on gpt-5.6-sol as a read-only advisor for a focused assumption, tradeoff, architecture question, security concern, difficult diagnosis, or high-risk decision. Use when the user requests a Codex second opinion or when another AI agent needs a bounded independent advisory pass; default reasoning effort to high unless the user explicitly supplies an effort.

Consult Claude Fable as a read-only advisor for a focused assumption, tradeoff, architecture question, security concern, difficult diagnosis, or high-risk decision. Use when the user requests a Fable second opinion or when another AI agent needs a bounded independent advisory pass; default effort to high unless the user explicitly supplies an effort.

Coordinate complex work as a controller-only orchestrator using cost- and risk-aware capability routing, explicit task dependencies and ownership, skeptical proof verification, and an independent review gate. Use when the user asks for delegated execution, parallel agents, controller-only operation, or rigorous multi-agent delivery.

Create or update an evidence-backed description for a pull request, merge request, or equivalent change proposal using the actual base, immutable branch diff, complete file coverage, verified tests and safety claims, and deterministic validation. Use when asked to draft, rewrite, standardize, or remotely update a proposal description.

Run an evidence-backed review of local files, staged or unstaged work, branches, commits, diff ranges, pull requests, merge requests, or equivalent remote proposals, with immutable diff coverage, calibrated tests, a validated decision, and optional explicitly authorized publication. Use when asked to review, audit, inspect, approve, request changes, comment on, or publish feedback for code changes.

Execute a software task through the complete bug-or-feature implementation, refinement, review, simplification, test-coverage, proposal, browser-proof, and demo-video workflow. Use when the user wants end-to-end delivery with no silently skipped phase, fresh proof after every change, a created or updated pull/merge request, and published video evidence.

Implement a new codebase capability end-to-end with frozen requirements and scope, risk-calibrated test-first delivery, behavior proof, validated evidence, and optional publication only when explicitly requested. Use for new screens, endpoints, integrations, data flows, commands, or other functional additions; do not use for correcting broken existing behavior, plan-only analysis, or code review.

Design, implement, and validate risk-based regression coverage across unit, component, integration, API/contract, and browser E2E layers, selecting the smallest reliable proof for each changed behavior. Use when asked to add tests, prove a bug fix, increase confidence or coverage, map acceptance criteria, or close backend or frontend test gaps.

Diagnose and repair broken existing behavior with exact reproduction, proven root cause, frozen scope, the smallest safe correction, calibrated regression proof, and validated evidence. Use for defects, regressions, crashes, incorrect results, failed user flows, or contract violations; do not use for new capabilities, plan-only analysis, or code review.

Stress-test and revise a technical strategy without implementing it, using an exact evidence baseline, fact and assumption ledgers, bounded loophole analysis, verification mapping, and a validated confidence decision. Use for implementation plans, debugging hypotheses, migrations, rollouts, releases, test strategies, or architecture approaches that need hardening before execution.

Simplify code introduced by a completed task while preserving observable behavior, public contracts, unrelated dirty work, and existing proof. Use after implementation already works when asked to reduce duplication, branching, state, indirection, speculative flexibility, or review complexity; do not use to add features, fix unknown defects, or perform broad cleanup.