Architect + SoD-review rigor for single-module critical-path changes that don't warrant a full /swarm but DO warrant more than bare /clean-code. Use when touching auth, sign-in, borrow, return, download, DRM fulfillment, audiobook playback, persistence migrations, or any code where a regression would hit users — regardless of LOC count. Invoke via "/rigorous-fix <task>" or when the user says "rigorous fix for X", "do this with SoD review", "critical path change to Y". For multi-module work, use /swarm. For non-critical bug fixes <50 LOC, /clean-code is sufficient.
Triage→dispatch→integrate→promote loop for multi-module Palace iOS work. Architect agent identifies touched modules, writes contract deltas, then parallel module-implementer subagents land changes against those contracts. forge-review and verify-pr.sh gate the integration. Use when a feature, refactor, or extraction touches ≥2 top-level Palace modules; for single-module work, do it directly. Invoke via "/swarm <task>" or when the user says "swarm this", "extract X via swarm", "run swarm for...", or asks to coordinate multi-module changes.
Run the SoD-required reviewers against the current ForgeOS changeset. Detects which gates are pending and require role-based reviews (architect, qa_test, blast_radius, etc.), spawns the matching subagent for each, and reports verdicts. Use when a PR is ready for review under harness governance and the author cannot self-approve. Invoke via "/forge-review" or when the user says "review my PR", "run the forge reviewers", "get architect/QA review on this changeset", or asks how to satisfy a blocked SoD gate.
Record a structured intent file (`.forgeos/intent/<name>.md`) BEFORE writing code for any change that adds ≥10 prod LOC under `Palace/`, touches a critical-path file (sign-in/auth/borrow/return/download/DRM/audiobooks/migrations/TPPNetworkExecutor/TPPNetworkResponder), or kicks off a `/swarm` or `/rigorous-fix` run. The intent file pins your Claims, Anti-claims, and Files-in-scope so the M1 contract-reconciliation gate (`scripts/check-contract-reconciliation.py`), the M1 intent-recorded gate (`scripts/check-intent-recorded.py`), and the universal pre-commit hook can verify the diff matches what you said you'd do. Invoke via "/intent <slug>" or whenever the user says "let's plan this first", "record the intent", "draft an intent file", "before we change code, write the plan", or whenever you (the agent) are about to start a ≥10 prod-LOC change.
Run a full release regression test — sets up workspace, runs automated tools, guides manual testing, generates report and Jira tickets. Use for release gates and QA cycles.
Audit staged or branch-scoped code for clean-code violations before commit/PR — DRY/duplication, dead code, premature abstraction, fluff tests, banned patterns (force unwraps, GCD where async exists, useless comments), and copy-paste drift. Use when about to commit code, when the harness pre-commit hook reports findings, when a subagent finishes an implementation, when the user says "clean code review", "check for duplication", "review my diff", "is this clean", or whenever you've just written non-trivial code and want a second pass. Auto-invoked by the pre-commit hook when static detectors find candidates.
Guide a QA tester through a manual release regression — setup, side-by-side testing, codebase-aware triage, report, and Jira tickets. No simdrive or maintainer-only tooling required. Use this when running a manual regression pass against a release candidate.