mkreyman

BLOCKS shortcuts like modifying dialyzer.ignore or .credo.exs excludes. Enforces fixing actual problems. Use when encountering ANY error, warning, or quality tool complaint.

MANDATORY pre-check before ANY response - ensures you use applicable skills and follow best practices. Use this skill FIRST before responding to any Elixir/Phoenix request.

PROHIBITS placeholder code, default values that mask missing data, and silent failures. Enforces fail-fast with loud errors. Use when implementing ANY function or data structure.

MANDATORY systematic debugging - trace to root cause before proposing fixes. NO random changes, NO symptom fixes, NO "try this". Use when debugging ANY error or issue.

MANDATORY for ANY feature or bugfix - write ExUnit test FIRST, watch it FAIL, then implement. NO exceptions. Use before writing any Elixir production code.

MANDATORY verification before claiming anything works. Run actual commands, read actual output, provide evidence. Use when claiming tests pass, build succeeds, or code works.

Create, manage, and safely run Ecto database migrations with proper rollback handling and best practices. Use when working with database schema changes, adding columns, or modifying constraints.

Run comprehensive Elixir quality checks (format, credo, dialyzer, tests) with proper error handling and reporting. Use when validating code quality, before commits, or preparing for deployment.

Use when working on loopctl's knowledge/retrieval product — the shared Knowledge Wiki, agent memory, the context retriever, hybrid (curated + RAG) search, the novelty/dedup gate, embeddings, conflict resolution, or KB curation permissions. Covers the three agent information surfaces and how to pick between them. Triggers on: knowledge, wiki, knowledge_create, knowledge_search, hybrid_search, novelty, dedup, proposal gate, embedding, vector search, conflict, curation, memory_remember, memory_recall, retrieve_, context retriever, entity, RAG, progressive disclosure, OKF.

Use when changing anything in loopctl's trust model — role requirements, the auth pipeline, story lifecycle transitions (report / review-complete / verify), capability tokens, dispatch lineage, WebAuthn/human-anchor gates, or the audit chain. Covers who may do what, why the self-* 409s are correct, and the capability/dispatch mechanics. Triggers on: chain of custody, self_report_blocked, self_review_blocked, self_verify_blocked, RequireRole, role hierarchy, capability token, dispatch, lineage, WebAuthn, human anchor, custody halt, superadmin, orchestrator, agent role, verify_story, report_story.

Use when touching data access, tenant scoping, RLS policies, migrations, or the repo topology in loopctl — anything that reads/writes tenant-scoped rows, adds a table, or runs heavy analytical/vector reads. Covers the three Ecto repos (Repo/AdminRepo/HeavyReadRepo), the SET LOCAL RLS mechanism, the transaction-owner invariant, and the pgbouncer statement_timeout gotcha. Triggers on: tenant_id, RLS, row level security, with_tenant, set_rls_context, SET LOCAL, app.current_tenant_id, BYPASSRLS, AdminRepo, HeavyReadRepo, heavy read, statement_timeout, pgbouncer, migration, new table.