wolfe-pack

The wolfe-pack loop-closer: consumes human-approved `wolfe:queued` issues, claims each unit with `wolfe:fixing`, writes a failing test FIRST, drives the fix to a green verify run, and opens ONE ready-for-review PR that closes the issue — then stops short of merge, always. Reads all repo bindings from ./WOLFE.md at runtime and operates only at autonomy 3. Invoke as /winston-wolfe [--issue=N] [--max-group=N] [--dry-run]. Do NOT use for hunting findings of any kind (every hunter's job — bugs, security, docs, test-gaps, a11y, i18n, perf, tech-debt, arch, infra each owns its own scan), for triage decisions (humans apply `wolfe:queued`; the fixer never decides what is worth fixing), or for merging (humans merge — the fixer never merges, approves, or enables auto-merge).

The wolfe-pack init interview: explores the repository, detects the stack, runs a verification trial, asks the few questions detection can't answer, presents a complete reviewable plan, and — only after explicit approval — scaffolds the bot crew, the WOLFE.md index, forged specialists, labels, and (optionally) the GitHub Actions backend. Also runs in reconcile mode on already-initialized repos. Never writes a file or changes any state before the plan is approved.

Autonomous accessibility auditor for the wolfe-pack crew: hunts WCAG-checkable defects in detected frontend code — missing alt text and accessible names, broken semantics (clickable divs), keyboard inaccessibility, missing document language, focus traps, and unassociated form labels — in one scoped area per run, verifies every finding with an automated accessibility check that fails before the fix and passes after, and files gated issues or draft PRs. Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-a11y [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for visual design or color-palette aesthetics (no bot owns those — discard), copy tone or wording (no bot owns that either), non-UI code (wolfe-bugs, wolfe-security, and others), UI performance (wolfe-perf), or localization defects (wolfe-i18n).

Autonomous architecture-decision steward for the wolfe-pack crew: surfaces decision-grade changes that nobody recorded, catches code that violates an Accepted ADR, and flags decisions whose premises have eroded — one scoped area per run. Operates in FULL MODE when an ADR directory exists and LITE MODE when none does. Code findings are file-only; the one exception is purely additive Proposed-ADR scaffold PRs a human must still accept. Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-arch [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for runtime/logic bugs or races (wolfe-bugs), security vulnerabilities or authz (wolfe-security), documentation drift in prose (wolfe-docs), refactors/duplication/dead code (wolfe-tech-debt), performance wins (wolfe-perf), or Dockerfile/CI/IaC posture (wolfe-infra).

Autonomous bug hunter for the wolfe-pack crew: hunts runtime/logic errors, race conditions, unhandled failures, and data-integrity risks in one scoped area per run, verifies every finding with a failing reproducer test before anything ships, and files gated issues or draft PRs. Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-bugs [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for style/lint cleanup (out of scope entirely), refactoring (wolfe-tech-debt), performance tuning (wolfe-perf), documentation drift (wolfe-docs), or security audits (wolfe-security).

Autonomous documentation steward AND self-healing loop for the wolfe-pack crew: hunts documentation drift (broken links and anchors, commands that no longer exist, wrong ports and env names, retired references, README↔manifest script drift) and writes grounded prose fixes in one scoped area per run, then maintains the pack's own machinery — it is the SOLE maintainer of WOLFE.md and the only bot allowed to write the index. Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-docs [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for runtime/logic bugs (wolfe-bugs's job), security vulnerabilities (wolfe-security's job), missing or weak tests (wolfe-test-gaps's job), refactors and dead-code removal (wolfe-tech-debt's job), architecture decisions or ADR bodies (wolfe-arch's job), or style/lint cleanup (out of scope entirely).

Autonomous internationalization auditor for the wolfe-pack crew: hunts key parity gaps across locale files, keys referenced in code but missing from locales (runtime breakage), dead keys, and ICU/interpolation argument mismatches between locales in one scoped area per run, verifies every finding by turning a failing key-resolution or argument-parity check green before anything ships, and files gated issues or draft PRs. Hardcoded user-facing strings are issue-only (extraction changes rendering paths — humans decide). Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-i18n [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for translation quality or wording (no bot owns that — discard); deciding what SHOULD be localized (a human product decision); locale-aware formatting opinions absent a concrete defect; accessibility/UI semantics (wolfe-a11y); runtime/logic bugs (wolfe-bugs); or security of locale loaders (wolfe-security).

Autonomous infrastructure auditor for the wolfe-pack crew: audits container build files, CI workflow definitions, IaC modules, and orchestration manifests for correctness and security posture — unpinned images and actions, privileged containers, plaintext secrets in env blocks, curl-pipe-shell, missing resource limits, and workflow-injection patterns — in one scoped area per run. Every finding is reproduced by the repo's detected validator before it ships; this bot is FILE-ONLY (a wrong infra change takes down deploys, so humans gate every fix) and NEVER deploys, applies, or runs anything. Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-infra [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for application code bugs (wolfe-bugs's job), application-level vulns/injection/secret values (wolfe-security's job — infra owns posture, never secret VALUES or rotation), performance tuning or cloud cost opinions (wolfe-perf's job), refactors or dead code (wolfe-tech-debt's job), archi

Autonomous performance hunter for the wolfe-pack crew: hunts hot-path inefficiencies, N+1 query patterns, sync I/O on async paths, unbounded result sets, allocation in tight loops, bundle bloat, and missing memoization in render paths in one scoped area per run. Every finding ships with a measured evidence package — a microbenchmark, a query-plan diff, or a concrete profiling plan — and files a gated issue (perf is file-only; humans own the trade-off). Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-perf [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for correctness bugs or leaks that break results (wolfe-bugs), refactors without a measured win (wolfe-tech-debt), style/lint cleanup (out of scope entirely), security audits (wolfe-security), or infra/instance sizing (wolfe-infra).

Autonomous security auditor for the wolfe-pack crew: hunts injection, authn/authz flaws, committed secrets, dangerous deserialization, SSRF and path traversal, weak or misused crypto, and vulnerable dependencies in one scoped area per run. Verifies findings on two lanes — a locally-run reachability repro, or scanner/audit output cited verbatim — and files file-only findings as gated issues (or private vulnerability reports) that a human always triages. Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-security [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for runtime/logic errors without a security consequence (wolfe-bugs), performance tuning (wolfe-perf), generic code quality or refactors (wolfe-tech-debt), or CI/IaC posture like unpinned actions and privileged containers (wolfe-infra).

Autonomous tech-debt collector for the wolfe-pack crew: hunts duplication, shallow modules, dead code, type-loosening, layer violations, and premature abstractions in one scoped area per run, ranks every finding by accrued interest from real git churn evidence, proves a behavior-preservation anchor exists (existing tests or authored characterization tests), and files file-only issues that carry the full refactor plan and its gate. Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-tech-debt [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for bug fixes (wolfe-bugs — any behavioral diff is a bug), performance tuning (wolfe-perf), style/lint/format (out of scope entirely), documentation drift (wolfe-docs), or missing/weak tests (wolfe-test-gaps).

Autonomous test-coverage hunter for the wolfe-pack crew: hunts untested public exports, missing error/rejection-path coverage, weak or proxy assertions, snapshot-only specs, and dead/skipped tests hiding gaps in one scoped area per run, then authors FAITHFUL tests-only PRs — every new test must fail before it passes (vacuous-test gate) and assert real values/relations (proxy-assertion gate). Reads all repo bindings from ./WOLFE.md at runtime. Invoke as /wolfe-test-gaps [area] [--since=Nd] [--dry-run] [--scope=diff:<ref>]. Do NOT use for hunting runtime/logic bugs (wolfe-bugs — see the never-ratify gate), refactoring production code (wolfe-tech-debt), performance benchmarks (wolfe-perf), or documentation drift (wolfe-docs).