achilles-orchestratorThe Achilles QA orchestrator. Runs one QA task end to end as a subagent of the main session: routes the ask to a workflow, dispatches inspector / author / reviewer / doc-writer experts with the Agent tool, keeps index-level state only, and hands back DONE, DONE_WITH_CONCERNS, NEEDS_CONTEXT or BLOCKE
agents-vs-agentsUse this skill for adversarial AI testing — red-teaming, guardrail verification, bias detection, prompt injection testing, and compliance auditing of any application with an AI component. Triggers on any mention of: "test AI guardrails", "adversarial testing", "red team the AI", "test for bias", "pr
bug-discoveryAdversarial probing of a live (or, in static mode, inferred) application for one journey, page set, or the whole app, loaded by an `author` dispatch. Triggers: "find bugs", "break the app", "bug hunt", "quality audit", "edge case testing", "exploratory testing", "find issues", and the adversarial pa
bug-reportUse when a defect, issue, or unexpected behavior is found during testing and needs to be reported in a defect tracking system (Jira). Use when asked to write, create, or file a bug ticket, defect report, or issue report. Loaded by a `doc-writer` dispatch; fields only the user knows come back as `NEE
contract-testingUse when writing, running or reviewing a test that calls a backend HTTP endpoint. Owns what every API interaction should assert: status codes, response schema, error envelope, headers. Activates on contract intent (contract or schema test, pact, OpenAPI or spec conformance, lock, verify or ensure th
database-testingUse whenever a test reads from or asserts against a SQL database (write, run, review, or just call). Owns the shape every database interaction should produce: row counts, cell values, column order, commit/rollback verification, and the DB as the oracle for a UI or API mutation. Activates on database
defect-confirmationIndependent reproduction of one suspected defect, loaded by an `achilles-reproducer` dispatch. Decides from the claim, the acceptance criteria, the live application and the source whether the defect is a bug, real only in the source, a spec ambiguity or intended, and returns that verdict as the stan
evidence-captureEvidence-first verification of one functional task, loaded by an `author` dispatch: compose one Steps-API spec with per-step screenshots, run it with video, trace, HAR and console capture, and freeze an immutable, redacted bundle under `tests/e2e/evidence/<slug>-<ts>/` that a human can open or attac
failure-diagnosisDiagnose failing Playwright tests through evidence-based triage, loaded by an `author` dispatch when a test fails in any mode (authoring, composing, probing, repair) or when the brief says "test is failing", "debug this", "why is this failing", "fix this test". Two entrypoints: a LOCAL failure (arti
journey-mappingThe expert half of journey mapping: how an `inspector` drives one app section and returns its flows (the per-section brief, the canonical section vocabulary, the gated-areas policy, the concurrency log), and how a `doc-writer` turns the cycle returns into the sentinel-bearing `tests/e2e/docs/journey
performance-testingk6 load, stress, spike, soak and breakpoint testing with SLO thresholds as the oracle, loaded by an `author` dispatch. Triggers: "load test", "performance test", "perf test", "k6", "stress test the endpoint", "check p95 under load", "latency under load", "how many concurrent users", "throughput", "r
qa-reportingStakeholder-facing output from an existing suite, loaded by a `doc-writer` dispatch. Two outputs sharing one render-and-inspect pipeline: the work summary deck ("generate a report", "QA report", "export a deck", "show what we've done", onboard Phase 8) and the test catalogue ("produce a test catalog
reviewThe one reviewer skill, loaded by every `reviewer` dispatch. Four modes selected by the brief: composition judge (Stage 4c: an independent, adversarial read of freshly composed specs against the scenario, the oracle ladder, the API reference and test-data feasibility), Stage B (the per-journey or ba
secrets-sweepExtract hardcoded credentials, API keys, PII-shape literals and app URLs out of a test suite into `.env`, loaded by an `author` dispatch. The final guardrail before a suite is published or treated as portable across environments; onboard Phase 7. Returns conform to the composer schema.
selector-developmentAdd one inert test attribute (data-testid or the project's detected convention) to a frontend element that has no stable selector, prove the change is inert with an eight-step guardrail pipeline (typecheck, unit, e2e, visual diff), and return the new selector. Loaded by an `author` dispatch on a sel
test-composerCompose the full test portfolio for one user journey (happy path, error states, edge cases, mobile, negative flows, data-lifecycle and visual variants), stabilise it, review it, verify coverage is exhaustive, and return. Loaded by an `author` dispatch whose brief names `journey=<j-id>` from a sentin
test-data-conventionsUse whenever a test's relationship to data is in question: what data a scenario needs, where it comes from, who creates it and who cleans it up. Triggers on "test data", "test data strategy", "hardcoded data", "dynamic test data", "seed data", "data cleanup", "data generation", "fixtures", "test fix