mili-ssm/agent-skills

HTTP contract rules for JSON and hypermedia endpoints - status-code choices, idempotency keys, RFC 9457 errors, pagination, caching, concurrency control, versioning and compatibility (Hyrum's law). Use when designing, reviewing or changing an API, endpoint, route, webhook or response format, including retry behavior and error formats; load it before designing or changing an endpoint so status codes, headers and compatibility calls follow its rules. Debugging a failing endpoint belongs to Superpowers systematic-debugging.

Runs the architect agent on a scope, design or goal and returns its C4 map, findings, options and ADR drafts.

Catalog of design principles, software laws, architectural styles, patterns, smells and evaluation methods (ISO 25010, ATAM, C4, fitness functions) for trade-off-aware recommendations. Use when evaluating or restructuring a codebase's architecture, module boundaries, layering or coupling, choosing between styles, or when one change ripples across many files; load it before recommending so each option names its quality attribute, trade-off and migration path.

Runs a feature briefing - brainstorm, parallel evidence gathering, options decision, plan and pre-mortem.

Maps the blast radius of a significant change - callers, contracts, data, docs, consumers, breaking changes, migration strategy (expand/contract, flags, strangler), rollout, rollback and point of no return. Use when planning or about to implement a rename, schema or API change, dependency swap, removal or boundary refactor, or when asked what a change will affect and how to roll it out safely; load it before planning a rename, schema, API, dependency or boundary change, because the plan inherits the blast radius.

Runs the consistency-auditor agent to find contradictions, ambiguities and stale decisions between code and docs.

Writes and audits concise architecture decision records (MADR or Y-statement) with alternatives, consequences, revisit triggers and a staleness rubric. Use when a technical decision is made, changed, questioned or needs writing up for future developers, when existing records need an audit, or to find the recorded rationale of an existing decision (check docs/adr first); a casual "why" question about a piece of code does not need it.

Runs the doc-curator agent to verify docs against code and compress them; edits only with 'apply'.

Checklists for accurate, dense repository documentation - Diátaxis, a density checklist and a compression workflow with before/after metrics. Use when writing, tightening, reviewing or updating a README, docs, ADRs, docstrings or a CLAUDE.md/AGENTS.md, including requests like "make this shorter", "tighten this section" or "this feels bloated" (not for general prose such as emails; how wording steers an LLM - prompts, skills, agents - belongs to prompt-engineering); load it before editing so cuts follow the density checklist and keep rationale and gotchas.

Orchestrates a briefing before building a new feature - brainstorming with the owner, parallel researcher, consistency-auditor and specialist subagents to contrast docs, code, requirements and external practice, then an options decision and an implementation plan stress-tested by pre-mortem. Use when asked how to approach or plan a new feature or significant change that needs evidence from several sources, or to compare ways of doing something before committing; small features go straight to Superpowers brainstorming or writing-plans. Load it before planning so evidence is gathered in parallel and the decision is recorded.

Reconstructs intent from code and docs, then interviews the owner in prioritized question rounds until no material ambiguity remains, recording answers as EARS requirements and ADRs. Use when requirements, constraints or the reasons behind existing code are unclear, undocumented, contradictory or possibly stale, for a whole project, a feature or a significant change; a new design with no existing code or docs belongs to Superpowers brainstorming.

Telemetry design method in four steps (map the request path, define wide-event and log fields, set SLIs and SLOs, alert on burn rate) with wide events (canonical log lines), structured logs, OpenTelemetry traces and metrics, RED/USE, cardinality and privacy rules, Python and Rust tooling. Use when deciding what to log, measure, trace or alert on, adding monitoring, dashboards or health checks, or preparing a service so production incidents can be debugged quickly; it designs the telemetry ahead of incidents, while diagnosing a live failure belongs to Superpowers systematic-debugging.

Forces delegation of a task to the cheapest suitable worker subagent and returns only its distilled result.

Runs the performance-analyst agent to baseline, profile and report measured improvements; optimizes only when asked.

Runs the pre-mortem agent on a plan or spec before execution and returns failure stories, risks, amendments and kill criteria.

Rules, labeled measured or prescribed, and workflows for writing text that LLMs read or act on - prompts and system prompts, skills, subagent and command definitions, worker briefs, tool and MCP descriptions, CLAUDE.md/AGENTS.md - and for shaping projects run mostly by coding agents, including cheap navigation of long files by outline and line range. Use when writing, improving or reviewing such text, when asking how to phrase an instruction so Claude follows it, why an agent ignores a rule, overtriggers, over-engineers or stops early, which model or effort a worker needs, or how to make a repository easier for agents; load it before answering so advice cites current Anthropic guidance and measured results instead of memory.

Runs the prompt-engineer agent to review skills, agents, commands, system prompts, tool descriptions or CLAUDE.md for LLM behavior; read-only.

Test design and test strategy toolkit - ISTQB techniques, SFDPOT and HICCUPPS heuristics, property-based, stateful and mutation testing, test doubles, test smells and flaky-test causes. Use when deciding which test cases to write for a function or feature, assessing a test suite or strategy, hunting coverage gaps, or fixing flaky tests, but not for the red-green-refactor loop itself (Superpowers test-driven-development); load it before answering so each case names its technique and the risk it covers.

Runs the qa-engineer agent to assess a test suite, design test cases, root-cause flaky tests or run mutation testing; writes tests only with 'write'.

Guides refactoring - restructuring code without changing its behavior - through characterization tests, small test-backed steps, Fowler's smell catalog, Tidy First, Mikado method, parallel change, branch by abstraction and strangler fig. Use when asked to refactor, restructure, clean up, split, rename, extract or untangle a function, module or legacy code, or to pay down tech debt; load it before proposing steps so each one names its smell and keeps tests green.

Conventions for Conventional Commits, SemVer bumps, Keep a Changelog entries and release checklists. Use when writing a commit message or changelog entry, choosing a version number, tagging or preparing a release; load it before proposing a version or tag so the bump follows the commit rules.

Offloads web research on one or more concrete questions to parallel researcher subagents and returns cited findings.

Runs the security-reviewer agent for a source-to-sink review, STRIDE threat model or design review.

Security review checklist that traces attacker input to dangerous sinks and rates exploitability - STRIDE, OWASP Top 10 and ASVS themes, and Python, Rust and htmx sinks with safe fixes. Use when asked whether code, an endpoint or a design is safe, or to check for vulnerabilities (injection, XSS, CSRF, SSRF, auth flaws), secrets, uploads or redirects; load it before answering so findings follow its source-sink format and severity scale.

Plans and designs a UI, page or user flow with the ux-principles design workflow and turns acceptance criteria into Playwright checks.

Plans, designs and reviews user interfaces and flows with Playwright as the verification loop - task and flow design, screen specs with all states, accessibility plans, acceptance criteria as Playwright checks, plus Nielsen heuristics, laws of UX, WCAG 2.2 AA, deceptive-pattern checks and a script capturing screenshots, axe-core violations and focus order. Use when planning or designing a page, form, component or user flow, defining its UX acceptance criteria, or auditing a running UI; load it before designing so every decision cites a heuristic, law or WCAG criterion.

Runs the ux-reviewer agent on a running UI or its templates and returns severity-rated usability and accessibility findings with evidence.

Research protocol that offloads web searching and reading to the cheap researcher subagent (Haiku) and returns cited lists, comparisons and conclusions on concrete questions - official docs, best practices, library options, version facts. Use when a task needs current external information, official documentation or community practice from several sources, a comparison of options or a version-sensitive answer (a single quick fact can use WebSearch directly); load it before searching so reading stays out of the main context and every claim comes back sourced.