A minimal catalog of my favourite skills for working with claude
Agent
Independent second-opinion agent that challenges triage recommendations or sprint plans. Reads the same artifacts as the planner/triager but approaches from a skeptical stance. Spawned by /jira:advisor.
Agent
Executes a single PLAN.md (one wave-plan) with atomic commits. Worktree-aware. Updates EXECUTION.md as it goes. Orchestrator spawns one executor per plan; multiple executors run in parallel within the same wave.
Agent
Validates that the executed sprint actually meets PLAN.md's Nyquist criteria. Fills coverage gaps by writing tests for any criterion not already covered. Implementation files are READ-ONLY. Spawned by /jira:execute after jira-executor finishes.
Agent
Adversarial peer reviewer for a drafted engineering paper. Independent of the drafting process — challenges claim strength, evidence quality, and promotional language before publish. Returns a structured verdict for stall detection. Spawned by /jira:paper after the draft is written.
Agent
Audits all -PLAN.md files in a sprint by working backward from the goal. Verifies multi-source coverage matrix, per-plan task budget, parallel-safety, and CONTEXT.md decision implementation. Independent of jira-planner. Returns structured issue list for stall detection. Spawned by /jira:plan after the planner finishes.
Agent
Reads BRIEF, RESEARCH, CONTEXT and produces one or more numbered PLAN files (01-PLAN.md, 02-PLAN.md, ...) grouped by execution wave. Each plan ≤ 3 tasks. Runs source-coverage audit and schema-push detection. Spawned by /jira:plan.
Agent
One focus area of sprint research. Spawned in parallel with sibling researchers by /jira:research. Reads the sprint BRIEF, investigates a single dimension (codebase, patterns, or external), and returns a focused findings document with confidence-tiered sources.
Agent
Reviews the diff for the current sprint (or current branch) and reports issues by severity. Doesn't fix — only reports. Spawned by /jira:review.
Agent
Goal-backward post-execution verifier. Asks "does the codebase as it stands now actually deliver the goal?" Distinct from jira-nyquist (test coverage) and jira-reviewer (diff audit). Spawned by /jira:execute after jira-nyquist passes.