roster

Audit code quality against KB-defined properties, invariants, and naming conventions.

Quality and compliance audit — combines code-quality and spec-compliance into one actionable report.

Periodic friction analysis — proposes new skills, deterministic tools, and adaptations.

Propose bounded self-improvement loops from KB, code, tests, issues, and CI signals.

Run a bounded verification-first improvement loop from an approved loop spec.

Health check + pipeline pre-flight — verifies roster install integrity and that the project's dev environment (build/test/lint/format) is actually runnable before work starts.

Formal verification gate — tool resolution via capability tag, roster re-runs coqchk/.itf replay directly, emits E0p/E0m/E0m-abstract evidence tier. Replaces the QA gate for --critical tasks.

Guided implementation — TDD, improve loop, sub-agents. Reads the plan, produces an impl brief.

Bootstrap a new project or onboard an existing project into the roster ecosystem.

Intake phase — transforms a task into a contractual brief validated by the human.

Root-cause investigation — analyzes a bug or unexpected behavior without modifying out-of-scope code.

Dual-voice decomposition — reads the intake brief, produces per-role sub-briefs.

Deterministic QA — quality gates, tmux matrix if TUI, blocked on review NO-GO.

Decompose a task into neutral research questions — blind research prep, task intent not revealed.

Blind documentarian research — reads questions only, never the task. Produces file:line grounded research with optional online prior-art scan.

Fix-first review with conditional specialists — produces a structured GO/NO-GO verdict.

Ship — conventional commits, rebase-merge, GitHub PR. Gated on review + QA go.

Implements skill-health approved improvements — skills, tools, adaptations, agents.

Formal spec phase — extends roster-spec output to produce a Rocq (.v) or Quint (.qnt) formal specification artifact. Runs after roster-spec, never instead of it.

Adversarial spec phase — derives user stories with concrete GWT scenarios, surfaces challenges, formalizes FR-NNN requirements, produces structured contract with runnable checks.

Critical-route triage — property elicitation, priority ordering, backend proposal, cost disclosure. Dispatched by roster-run when --critical is chosen. Checkpoints to briefs/<slug>-formal-triage.md.

Generic, propose-only upgrader for any roster-contract skill or pack. Mines evidence-graded signal, routes each change at its natural altitude, and gates it with a git-diff-enforced leak scan, the target's own validator, and a human-validation quiz the skill runs itself (altitude-independent). Propose-only — lands nothing; a human reviews and merges the diff. Maintainer-invoked only — never auto-discovered.

Translates a validated plan JSON into a CWR workflow file using the matching mode template (mechanical template-fill — template steps are copied verbatim). Triggered automatically by roster-run after plan COMPLETED when no workflow file exists yet.

Manage the installed agent team — build (apply proposal), review (audit gaps), or run (execute a task through the pipeline).

Pipeline entry point — detects context and routes to the right skill.

Generate or edit images via Codex CLI — with prompt refinement, vision validation, retry loop, and error handling.

Use when the user invokes /recruit, $recruit, recruit update, or asks to assemble, audit, update, or govern an agent team using mathiasbourgoin/roster.

Audit KB for ambiguity — undefined terms, vague requirements, contradictions, stale content.

Code quality and architecture guardian focused on structural regressions, duplication, and maintainability risks.

Apply git workflow conventions — commits, branches, PRs.

Meta-auditor — validate the KB harness itself (structure, auditors, rules coherence, feedback loops).

Executes scoped feature/fix tasks in isolated worktrees with deterministic verification before handoff.

Bootstraps and maintains project knowledge bases as source-of-truth artifacts for specs, properties, and architecture.

Audit, clean, reorg, and migrate an existing KB to the current schema — idempotent, human-gated at each phase.

Update knowledge base — sync KB files with recent code changes without weakening specs.

Takes a validated research brief and decomposes it into compressed, verified sub-briefs for each execution agent. Runs with a fresh context — no research history.

Verifies implemented behavior through deterministic test execution and focused scenario checks.

Performs structured code review focused on correctness, security, and regression risk.

Discover, inspect, and interactively set tunables across installed roster agents — with guided dependency installation for optional features.

Reverse-engineer what existing code actually does into a structured, evidence-tiered inferred spec artifact (specs/<slug>-inferred.md). Mines tests, code, documentation, and git history. Labels every claim [E1]/[E2]/[E3] by evidence strength. Produces GWT scenarios, FR-NNN requirements, and surfaces doc-drift conflicts.