loyal-opposition-hygiene-assessment

name	loyal-opposition-hygiene-assessment
description	Use when Loyal Opposition should assess GT-KB project hygiene, classify drift, and prepare a Prime Builder action plan without mutating project files.

Loyal Opposition Hygiene Assessment

Purpose

A Loyal Opposition advisory orchestration skill that synthesizes the existing narrower hygiene surfaces (structural hygiene, Deliberation Archive health, wrap-scan, harness parity) into a single Prime-facing action plan. The skill does not replace the existing skills; it orchestrates them and produces a standard advisory report.

This skill is read-only and advisory by default. It assesses; it does not mutate. Prime Builder owns implementation; Loyal Opposition owns standards and enforcement.

v1 Scope

This skill is v1 manual only. The following are explicitly out of v1 scope and must NOT be added without a separate owner-approved bridge proposal:

• A verify <report-or-phase> mode (deferred).
• A startup-pulse mode (lightweight session-start surface; deferred).
• A ::hygiene command surface or gt hygiene scan CLI (deferred).
• Scheduler / cron integration (deferred).
• Dashboard integration (deferred).
• Promotion of parity_class from baseline to required (deferred until v1 stabilizes).
• Any mutation surface attached to the skill itself.

Modes (v1)

overview

Broad read-only assessment across all nine hygiene phases. Produces one timestamped advisory report.

phase <id>

Targeted assessment for one hygiene phase by its numeric ID (1-9). Produces one timestamped advisory report scoped to that phase.

There are no other modes in v1.

Required Inputs

Read the smallest relevant set:

• .claude/rules/canonical-terminology.md
• .claude/rules/operating-model.md
• .claude/skills/structural-hygiene-review/SKILL.md
• .claude/skills/check-deliberations/SKILL.md
• .claude/skills/kb-session-wrap-scan/SKILL.md
• .claude/skills/harness-parity-review/SKILL.md
• config/agent-control/harness-capability-registry.toml
• config/agent-control/system-interface-map.toml
• Bridge dispatcher status/health CLI and TAFE-backed bridge-state surfaces (when bridge hygiene or bridge actionability is in scope). Do not consult or recreate aggregate queue artifacts as live bridge sources.
• python -m groundtruth_kb backlog list (only when backlog/work-item hygiene is in scope; reads MemBase work_items, the canonical backlog authority per ADR-STANDING-BACKLOG-DB-AUTHORITY-001 v2)
• Relevant docs, scripts, generated artifacts, tests, and archive paths for the selected hygiene phase

Nine Hygiene Phases

Phase	LO assessment responsibility	Prime execution expectation
1. DA harvest	identify missing or candidate Deliberation Archive capture and coverage risks	harvest / capture through the governed scripts/harvest_session_deliberations.py path
2. Branch cleanup	identify stale branches / worktree noise; require dry-run evidence before any cleanup	perform branch cleanup only with safe scope, never git push --force to main/master
3. Bridge double-version	audit bridge/ parser behavior, aggregate-queue assumptions, and version ambiguity before any fix	repair only if verified defect exists; never edit a filed bridge file in place
4. REPOSITORY-STRUCTURE.md	identify current structure / source-of-truth gaps	draft or update documentation
5. Terminology drift	map drift against .claude/rules/canonical-terminology.md	update docs / schema / API / tests as needed
6. Duplicate specs	identify candidate duplicate specifications and authority conflicts	merge / retire only through the governed spec process (formal-artifact approval)
7. Gitignore + scripts triage	identify noisy / generated / unowned artifacts and stale scripts	mechanical cleanup with verification; never script deletion without bridge-tracked approval
8. Naming consistency	identify invasive rename clusters and risk boundaries	execute in narrow slices after other phases stabilize
9. Smart-poller residual	identify retired smart-poller and OS-poller residue and live-dependency risk	remove or relabel residuals only after confirming archival status

Workflow

1. Determine mode (overview or phase <id>).
2. Read the smallest relevant input set for the chosen mode.
3. For each in-scope hygiene phase, run the relevant existing skill / scanner as a read-only orchestration step. Examples:
   • Phase 1: python scripts/check_deliberation_archive_health.py or the equivalent skill.
   • Phase 4-5: orchestrate structural-hygiene-review and check-deliberations outputs.
   • Phase 9: python scripts/check_harness_parity.py --all --markdown for adapter / parity residuals.
4. Synthesize findings into the standard report format (next section).
5. Write the report to independent-progress-assessments/CODEX-INSIGHT-DROPBOX/INSIGHTS-<UTC-timestamp>-LO-HYGIENE-ASSESSMENT.md.
6. Do not mutate any project file outside the report destination. No bridge edits, no MemBase mutation, no script deletion, no role / role-map mutation, no branch operation, no formal-artifact creation by this skill.

Required Report Format

Every generated advisory report MUST include the following sections:

1. Claim - one-sentence summary of the hygiene state assessed.
2. Scope assessed - which phase(s) were covered; which input files were read.
3. Evidence paths and line references - concrete file paths + line numbers for every cited finding.
4. Severity-ranked findings - P0 / P1 / P2 / P3 with rationale for each level.
5. Phase-ranked action plan - findings grouped by hygiene phase (1-9) with an explicit ordering recommendation.
6. Prime Builder implementation sequence - ordered, executable Prime work packets per finding.
7. Peer Prime delegation candidates - mechanical findings safe for Prime-mode harness delegation.
8. LO verification plan - Loyal Opposition follow-up after Prime implementation.
9. Explicit do-not-touch / deferred areas - boundaries the report assesses but recommends NOT acting on this round.
10. Owner decisions required - listed only when genuinely blocking; otherwise stated as "None" with rationale.
11. Residual risk - unresolved or de-prioritized risks after Prime executes the action plan.

Action Classification

Every actionable finding in the report MUST be classified as exactly one of:

• prime-action - Prime Builder should execute directly.
• peer-prime-candidate - Prime Builder may delegate mechanical execution to another Prime-mode harness.
• lo-verification - Loyal Opposition should verify after Prime implementation.

A finding with no clear ownership is a defect in the assessment; either refine the finding or omit it.

Severity Scale

• P0 - competing authority or source-of-truth conflict; active misdirection.
• P1 - active-looking obsolete artifact likely to mislead agents; governance drift.
• P2 - terminology drift across docs / schema / API / tests; capability overclaim.
• P3 - clutter or weak naming that reduces intuitiveness; terminology noise.
• P4 - historical context preserved without remediation.

Out-of-Scope Actions (LO Standards Boundary)

This skill is read-only advisory. The following actions are explicitly OUT OF SCOPE for ordinary LO assessment performed under this skill:

• Branch deletion or destructive git cleanup.
• Direct formal-artifact mutation (GOV / SPEC / ADR / DCL / PB inserts or new versions; Deliberation Archive inserts).
• Spec merge / retire operations.
• Script deletion.
• Broad rename / naming-consistency rewrites.
• Role-assignment or bridge-status mutation, except existing bridge-function repair authority that already lives elsewhere.
• Any write to groundtruth.db or .groundtruth/formal-artifact-approvals/.
• Any write to aggregate queue artifacts or bridge/*.md outside the standard Loyal Opposition bridge-review surface (which is NOT this skill). Aggregate queue artifacts are not canonical bridge-state authority and must not be recreated.

If the assessment finds a defect that requires mutation, the report MUST list it as a prime-action or peer-prime-candidate and stop there. Mutation is filed through standard Prime channels (bridge proposal -> Codex GO -> implementation -> post-impl -> VERIFIED), not through this skill.

Report Destination

Reports go ONLY to independent-progress-assessments/CODEX-INSIGHT-DROPBOX/ per .claude/rules/loyal-opposition.md storage convention. Naming convention:

INSIGHTS-<UTC-timestamp>-LO-HYGIENE-ASSESSMENT-<overview|phase-N>.md

Example: INSIGHTS-2026-05-29-13-45-LO-HYGIENE-ASSESSMENT-overview.md.

Relationship to Existing Skills

This skill orchestrates rather than replaces:

• structural-hygiene-review - structural drift, naming, artifact authority, glossary alignment.
• check-deliberations - Deliberation Archive population, linkage, conflict, redaction, duplicates.
• kb-session-wrap-scan - W1 (transcript snapshot) + W2 (cross-artifact consistency) scanners.
• harness-parity-review - cross-harness skill / adapter parity drift.
• arch-audit - architecture compliance audit when a phase touches ADR / DCL.

The orchestration value is in synthesizing those outputs into a single Prime-facing action plan with consistent classification (prime-action / peer-prime-candidate / lo-verification) and a phase-indexed ordering recommendation.

Future Slices (Not v1)

After v1 has produced one or two manual reports and the signal/noise level is calibrated, separate bridge proposals may add (in order):

1. verify <report-or-phase> mode for LO post-implementation verification.
2. startup-pulse mode that surfaces last hygiene report age, open P0/P1 count, and recommended next phase in session startup.
3. Scheduled report (weekly / pre-release / post-major-churn); report-only, no mutation.
4. ::hygiene or gt hygiene scan command surface aliasing this skill.
5. Promotion of parity_class from baseline to required.

Each of those is a separate bridge proposal with its own owner approval where applicable.

Source Authority

• Source advisory: DELIB-1473 (Loyal Opposition Advisory: LO Hygiene Assessment Skill).
• Disposition thread: bridge/gtkb-lo-hygiene-assessment-skill-advisory-disposition-002.md (GO at -002; VERIFIED at -004).
• Build thread: bridge/gtkb-lo-hygiene-assessment-skill-build-005.md (GO; this skill source).
• Owner authorization: DELIB-S350-BATCH4-FOUR-PROJECT-AUTHORIZATIONS via PAUTH-PROJECT-GTKB-LO-ADVISORY-INTAKE-LO-ADVISORY-INTAKE-PARALLEL-BATCH (covers WI-3303).

Copyright 2026 Remaker Digital, a DBA of VanDusen and Palmeter, LLC. All rights reserved.