Claude Skills by ReinaMacCredy

Design or review CLIs so Codex and other coding agents can run them reliably in terminals and automation: non-interactive flags, layered --help with real examples, stdin and pipeline support, fast actionable errors, idempotency, dry-run, confirmation bypass flags, and predictable command structure. Use when building a CLI, adding subcommands, writing help text, improving terminal automation, or making command UX safe for headless agent use.

Implement Mission Control domain logic, runtime-state persistence, recovery orchestration, and cross-CLI adapters in the Maestro CLI

Implement Mission Control CLI commands, Mission Control snapshot/TUI behavior, and user-visible recovery/operator workflows in Maestro

Branch: `harness-os` (PR #64) Generated: 2026-05-16 Owner: Plan agent (opus) per master-plan loop

- Subagent-only pattern (Task tool) as cross-platform common denominator -- no Agent Teams dependency - Planner subagent uses AskUserQuestion directly (general-purpose type, not Plan type) for user interviews - Plan handoff via draft file (.maestro/drafts/) avoids Task return value truncation for large plans - Tilde fences (~~~) for nested code blocks inside markdown skill files

Interview-driven product-spec authoring for maestro. Runs the grill protocol from ADR-0016 — walks the decision tree one branch at a time, challenges user language against CONTEXT.md and committed ADRs, cross-references claims against code when verifiable, updates CONTEXT.md inline as terms resolve, and offers ADRs sparingly. Produces a committed product-spec markdown at `.maestro/specs/<slug>.md` ready for `maestro task from-spec`. Use when the user wants to author or refine a spec before pl...

Read handoff envelopes left by previous agents at the start of a session or when picking up a task in a maestro-initialized project. Use whenever you suspect a prior session left context behind — task was claimed by another agent, a block was raised, or a new shell is taking over an unfinished workflow.

Turn an approved heavy-mode product-spec into an executable mission with child tasks. Use after `maestro-design` has produced a `mode: heavy` spec, or when a single task has grown big enough that it should be decomposed into a multi-PR batch. Persists the mission to `.maestro/missions/missions.jsonl` and the child tasks to `.maestro/tasks/tasks.jsonl`.

Set up a repository as a long-running agent harness. Use when a project needs Maestro-owned context docs, evidence-first onboarding, root AGENTS.md guidance, language style guides, host-runtime session hooks, drift detection, or a setup report before implementation work begins.

Use at the start of any multi-step work in a maestro-initialized project, and throughout task execution. Claim one task at a time, iterate through the verify → block / ship loop. `claim` and `block` automatically emit a handoff envelope to `.maestro/handoffs/<hnd-...>.json`; see `maestro-handoff` for the read side. Auto-invokes whenever a `.maestro/` directory is present in the current working tree or an ancestor.

The canonical verification protocol for any task in a maestro project. Documents witness levels, Trust Verifier scope, ProofMap, plan-check, verdict semantics, cost-budget monitoring, AI Reviewer protocol (Rule 1 veto-only), and threat-model production. Cross-referenced by `maestro-task` and `maestro-mission`. Read this skill when starting a non-trivial task or before declaring completion.

AI agent workflow with interview-driven planning and team-based execution. Use /design to start planning, /work to execute.

Scaffold a blank plan with required sections in .maestro/plans/. Use /plan-template <name> to create a new plan file.

Discovers host project conventions from CLAUDE.md, package.json, linter configs, and other configuration files. Use when agents need to understand the target project's coding standards.

Template for creating skills. Use when writing new code, reviewing code, or when the skill's specific domain applies. Replace this description with specifics.