Health Checks (Inline Orchestration) — 5 modes
You ARE the orchestrator. Execute the workflow directly without spawning a
separate orchestrator agent. Pick the mode, open its full workflow in
references/modes/<mode>.md, use worker prompts from
references/worker-prompts/<mode>.md.
Mode table
| Mode |
Command |
Detect worker |
Fix worker |
Priorities |
Report |
Changes log |
Fixes artifact |
| bug |
/health-bugs |
bug-hunter |
bug-fixer |
critical → high → medium → low |
bug-hunting-report.md |
bug-changes.json |
bug-fixes-implemented.md |
| cleanup |
/health-cleanup |
dead-code-hunter |
dead-code-remover |
critical → high → medium → low |
dead-code-report.md |
cleanup-changes.json |
dead-code-cleanup-summary.md |
| deps |
/health-deps |
dependency-auditor |
dependency-updater |
critical → high → medium → low |
dependency-scan-report.md |
deps-changes.json |
dependency-updates-implemented.md |
| reuse |
/health-reuse |
reuse-hunter |
reuse-fixer |
high → medium → low (no critical) |
reuse-hunting-report.md |
reuse-changes.json |
reuse-consolidation-implemented.md |
| security |
/health-security |
security-scanner |
vulnerability-fixer |
critical → high → medium → low |
security-scan-report.md |
security-changes.json |
security-fixes-implemented.md |
All 10 workers live in agents/health/workers/ (verified 2026-07-18).
Shared workflow skeleton (identical in every mode)
Detection → Validate → Fix/Remove/Update/Consolidate by Priority → Verify → Repeat if needed
Max iterations: 3
- Phase 1 — Pre-flight:
mkdir -p .tmp/current/{plans,changes,backups};
validate environment (package.json, type-check + build scripts exist);
initialize TodoWrite with one item per priority + detection + verification.
- Phase 2 — Detection: invoke the mode's detect worker via Task tool
(prompt in the mode file); read the report, parse counts by priority;
zero findings → skip to Final Summary.
- Phase 3 — Quality Gate (Detection):
pnpm type-check && pnpm build
inline; fail → report to user, exit.
- Phase 4 — Fixing Loop: for each priority — invoke the mode's fix worker
via Task tool (backup each file, log to the mode's changes json, produce the
fixes artifact), then inline quality gate; FAIL → report error, suggest
rollback, exit; PASS → next priority.
- Phase 5 — Verification: re-invoke the detect worker in verification mode,
compare with the previous report. remaining == 0 → Final Summary;
iteration < 3 and remaining > 0 → back to Phase 2; iteration >= 3 → Final
Summary with remaining items.
- Phase 6 — Final Summary: iterations {count}/3, status SUCCESS/PARTIAL,
found/fixed/remaining totals, per-priority breakdown, validation status,
artifacts list (exact templates — in the mode files).
Error handling (all modes): quality gate fails → rollback available from the
mode's changes log + .tmp/current/backups/ (or rollback-changes skill);
worker fails → report error, suggest manual intervention, exit workflow.
Per-mode deltas (details in references/modes/.md)
- bug — scan: type-check/build, security vulns, dead code, debug statements.
- cleanup — scan: unused imports/exports, commented blocks, unreachable
code, debug statements, unused variables/functions/dependencies; Knip.
- deps — extra pre-flight: lockfile must exist (pnpm-lock.yaml,
package-lock.json, yarn.lock); updater works ONE dependency at a time with
validate-after-each and per-dependency rollback; rollback additionally
restores package.json + lockfile and runs
pnpm install.
- reuse — priorities high/medium/low only (no critical); extra section
«Duplication Categories» + Single Source of Truth pattern (canonical
packages/shared-types/src/, re-export export * from '@package/shared-types/{module}', NEVER copy code between packages).
- security — scan: SQLi, XSS, auth/authz, RLS policy violations, hardcoded
secrets, insecure dependencies.
Why inline (vs old orchestrator agent)
| Old (Orchestrator Agent) |
New (Inline Skill) |
| 9+ orchestrator calls |
0 orchestrator calls |
| ~1400 lines (cmd + agent) |
~150 lines |
| Context reload each call |
Single session context |
| Plan files for each phase |
Direct execution |
| ~10,000+ tokens overhead |
~500 tokens |
References
references/modes/{bug,cleanup,deps,reuse,security}.md — full original
workflows, verbatim (formerly the 5 standalone skills).references/worker-prompts/{bug,cleanup,deps,reuse,security}.md — detailed
worker prompts (bug = original; other 4 written at merge time, closing the
links that used to point at a single non-existent worker-prompts file).
1---2name: health-inline3description: Codebase health инлайн (сам оркестратор): детекция→фикс→верификация. Триггеры: /health-bugs, /health-cleanup, /health-deps, /health-reuse, /health-security.4---56# Health Checks (Inline Orchestration) — 5 modes78You ARE the orchestrator. Execute the workflow directly without spawning a9separate orchestrator agent. Pick the mode, open its full workflow in10`references/modes/<mode>.md`, use worker prompts from11`references/worker-prompts/<mode>.md`.1213## Mode table1415| Mode | Command | Detect worker | Fix worker | Priorities | Report | Changes log | Fixes artifact |16|---|---|---|---|---|---|---|---|17| bug | /health-bugs | bug-hunter | bug-fixer | critical → high → medium → low | bug-hunting-report.md | bug-changes.json | bug-fixes-implemented.md |18| cleanup | /health-cleanup | dead-code-hunter | dead-code-remover | critical → high → medium → low | dead-code-report.md | cleanup-changes.json | dead-code-cleanup-summary.md |19| deps | /health-deps | dependency-auditor | dependency-updater | critical → high → medium → low | dependency-scan-report.md | deps-changes.json | dependency-updates-implemented.md |20| reuse | /health-reuse | reuse-hunter | reuse-fixer | high → medium → low (no critical) | reuse-hunting-report.md | reuse-changes.json | reuse-consolidation-implemented.md |21| security | /health-security | security-scanner | vulnerability-fixer | critical → high → medium → low | security-scan-report.md | security-changes.json | security-fixes-implemented.md |2223All 10 workers live in `agents/health/workers/` (verified 2026-07-18).2425## Shared workflow skeleton (identical in every mode)2627```28Detection → Validate → Fix/Remove/Update/Consolidate by Priority → Verify → Repeat if needed29```3031**Max iterations**: 332331. **Phase 1 — Pre-flight**: `mkdir -p .tmp/current/{plans,changes,backups}`;34 validate environment (`package.json`, `type-check` + `build` scripts exist);35 initialize TodoWrite with one item per priority + detection + verification.362. **Phase 2 — Detection**: invoke the mode's detect worker via Task tool37 (prompt in the mode file); read the report, parse counts by priority;38 zero findings → skip to Final Summary.393. **Phase 3 — Quality Gate (Detection)**: `pnpm type-check && pnpm build`40 inline; fail → report to user, exit.414. **Phase 4 — Fixing Loop**: for each priority — invoke the mode's fix worker42 via Task tool (backup each file, log to the mode's changes json, produce the43 fixes artifact), then inline quality gate; FAIL → report error, suggest44 rollback, exit; PASS → next priority.455. **Phase 5 — Verification**: re-invoke the detect worker in verification mode,46 compare with the previous report. remaining == 0 → Final Summary;47 iteration < 3 and remaining > 0 → back to Phase 2; iteration >= 3 → Final48 Summary with remaining items.496. **Phase 6 — Final Summary**: iterations {count}/3, status SUCCESS/PARTIAL,50 found/fixed/remaining totals, per-priority breakdown, validation status,51 artifacts list (exact templates — in the mode files).5253**Error handling** (all modes): quality gate fails → rollback available from the54mode's changes log + `.tmp/current/backups/` (or `rollback-changes` skill);55worker fails → report error, suggest manual intervention, exit workflow.5657## Per-mode deltas (details in references/modes/<mode>.md)5859- **bug** — scan: type-check/build, security vulns, dead code, debug statements.60- **cleanup** — scan: unused imports/exports, commented blocks, unreachable61 code, debug statements, unused variables/functions/dependencies; Knip.62- **deps** — extra pre-flight: lockfile must exist (pnpm-lock.yaml,63 package-lock.json, yarn.lock); updater works ONE dependency at a time with64 validate-after-each and per-dependency rollback; rollback additionally65 restores package.json + lockfile and runs `pnpm install`.66- **reuse** — priorities high/medium/low only (no critical); extra section67 «Duplication Categories» + Single Source of Truth pattern (canonical68 `packages/shared-types/src/`, re-export `export * from69 '@package/shared-types/{module}'`, NEVER copy code between packages).70- **security** — scan: SQLi, XSS, auth/authz, RLS policy violations, hardcoded71 secrets, insecure dependencies.7273## Why inline (vs old orchestrator agent)7475| Old (Orchestrator Agent) | New (Inline Skill) |76|--------------------------|-------------------|77| 9+ orchestrator calls | 0 orchestrator calls |78| ~1400 lines (cmd + agent) | ~150 lines |79| Context reload each call | Single session context |80| Plan files for each phase | Direct execution |81| ~10,000+ tokens overhead | ~500 tokens |8283## References8485- `references/modes/{bug,cleanup,deps,reuse,security}.md` — full original86 workflows, verbatim (formerly the 5 standalone skills).87- `references/worker-prompts/{bug,cleanup,deps,reuse,security}.md` — detailed88 worker prompts (bug = original; other 4 written at merge time, closing the89 links that used to point at a single non-existent worker-prompts file).
JHamidun