2475 Grooming 2026 02 21 F58716b3

Backlog Grooming Report

Date: 2026-02-21 Items groomed: 39 active (1 DONE skipped) Arguments: all Groomer agents launched: 10 (2 waves of 5) + 1 single-item Session branch: claude/bulk-backlog-grooming-LIQDi

Summary

Individual Manifests

hallucination-detector: Fix dead backtick evidence marker and multi-occurrence "because" bug (P1)

RT-ICA: APPROVED — both bugs VERIFIED from source, 1 assumption (test file location)

Fact-check: 2/2 VERIFIED from direct source read (primary source: plugins/hallucination-detector/scripts/hallucination-audit-stop.js)

Bug 1 (lines 111-114 vs 127-141): stripLowSignalRegions() removes inline code spans before hasEvidenceNearby() is called. EVIDENCE_MARKERS[0] (/\[^`]+`/) can never match in the already-stripped haystack. Fix: pass original unstripped text tohasEvidenceNearby()`.

Bug 2 (lines 223-229, 306-332): All phrase loops use lower.indexOf(phrase) which returns only the first occurrence index. Subsequent occurrences of the same phrase are never evaluated. Fix: replace indexOf with while/matchAll loop.

Skills: none specifically — standard JS fix Agents: @javascript-pro (matchAll pattern), @code-review Prior work: plugins/hallucination-detector/README.md, plugins/hallucination-detector/.claude-plugin/plugin.json Missing: No test file for hallucination-audit-stop.js — create test fixtures in plugins/hallucination-detector/tests/ Effort: Small (2 focused logic changes in one file)

Suggested first steps:

1. Create test fixtures with backtick-evidence + multi-occurrence "because" scenarios
2. Fix Bug 1: thread original text into hasEvidenceNearby() calls alongside stripped haystack
3. Fix Bug 2: replace indexOf() with matchAll() loop in speculationPhrases and causalityPhrases loops
4. Run uv run prek run --files plugins/hallucination-detector/scripts/hallucination-audit-stop.js
5. Validate plugin: uv run plugins/plugin-creator/scripts/plugin_validator.py plugins/hallucination-detector/

plugin-validator UX and coverage gaps (P1)

RT-ICA: APPROVED — 0 blockers

Key discoveries (groomer agent ae49af7):

• FileType enum already has HOOK_CONFIG variant (lines 318-329) — not missing as backlog implied
• test_hook_validator.py already exists — sub-issue 3 (hook support) has test infrastructure ready
• DescriptionValidator already has file_type awareness at line 1981 — sub-issue 2 smaller than expected
• 19 test files available (backlog stated 12)
• QA report at plugins/plugin-creator/planning/plugin-validator-qa-report.md maps all 4 sub-issues

Recommended order: Sub-issue 4 (dead code removal, lines 904-911) → Sub-issue 1 (UX, lines 2928-2931) → Sub-issue 3 (hook support) → Sub-issue 2 (command SK005)

Skills: plugin-creator:claude-hooks-reference-2026, python3-development Agents: @refactor-executor, @plugin-assessor Prior work: plan/tasks-2-validator-ux-coverage.md, planning/plugin-validator-qa-report.md

Validate and verify orchestrator-discipline plugin (P1)

RT-ICA: APPROVED — 0 external blockers; internal blockers are planned tasks

Key discoveries (groomer agent a37eae0):

• Plugin currently FAILS claude plugin validate due to invalid rules field in plugin.json
• SKILL.md missing user-invocable: true frontmatter — skill activation broken
• Hook gap: directory-targeted Grep calls (e.g., path: "src/") are NOT detected
• Plan at plan/tasks-4-validate-orchestrator-discipline.md is comprehensive with 4 tasks (T1-T4), architectural decisions pre-made, 5 verification test commands specified

Task execution order: T1 (fix plugin.json invalid rules field) + T2 (add directory detection to hook) + T3 (add user-invocable: true) → T4 (convergence validation)

Skills: plugin-creator:claude-plugins-reference-2026 Agents: @general-purpose (per plan task assignments) Prior work: plan/tasks-4-validate-orchestrator-discipline.md

Add PR003/PR004 test coverage (P2)

RT-ICA: APPROVED — 0 blockers

Key discoveries (groomer agent abe9d6c):

• PR003 = info severity (not warning/error) — tests must check result.info list
• PR004 = warning severity — tests must check result.warnings list
• Helper functions _make_plugin() (lines 32-64) provide complete test setup pattern
• Existing PR001/PR002 test classes (lines 267-543) serve as templates
• Git metadata extraction may require monkeypatch for PR004 tests

Target file: plugins/plugin-creator/tests/test_plugin_registration_validator.py New classes: TestMissingMetadata (PR003) and TestRepositoryMismatch (PR004)

SAM Extension — Integrate ARL General Theory (P1)

RT-ICA: APPROVED — 0 blockers

Key discoveries (groomer agent af51cae):

• Path correction: ARL references are at plugins/plugin-creator/skills/arl/references/ NOT skills/assessor/references/ARL/ (backlog path is wrong)
• All 7 principles fully documented in synthesis-general-theory.md (173 lines) with evidence from 6 frameworks
• synthesis-arl-applicable.md provides mapping template (R1-R10 → framework patterns)
• Target files are in external repo bitflight-devops/stateless-agent-methodology (confirmed)
• Blocks 5 related P1 SAM gap items

Requires: Cross-repo work in bitflight-devops/stateless-agent-methodology

ARL Skill Development (P1)

RT-ICA: APPROVED — 0 blockers; design question before starting

Key discoveries (groomer agent a37be14):

• ARL SKILL.md scaffold already exists at plugins/plugin-creator/skills/arl/SKILL.md
• R1-R10 gates fully specified in synthesis-arl-applicable.md (305 lines)
• Design question: Reference skill (documents gates) vs. Orchestration skill (executes ARL loop)?
• Backlog scope says "logical process design" → reference skill, but assessor SKILL.md shows orchestration pattern
• Soft dependency on SAM Extension (not blocking)

Meta-Process Capture — Expert Panel Dataset Builder (P1)

RT-ICA: APPROVED — design decision needed before starting

Key discoveries (groomer agent a7d0457):

• .claude/commands/arl-expert-panel.md already exists (100 lines) — process is operational
• ARL-agent-instructions.md at plugins/plugin-creator/skills/arl/references/ (440 lines) — complete process specification
• Design decision: Skill format (.claude/skills/expert-panel-methodology/SKILL.md) vs. reference document
• Blocks SAM Extension and ARL Skill Development items

SAM: Parser regex false positive (P2)

RT-ICA: APPROVED — soft blockers

Key discoveries (groomer agent aea3b2f):

• File: plugins/python3-development/skills/implementation-manager/scripts/implementation_manager.py line 645
• Regex: ^#{2,3}\s+Task:?\s+([A-Za-z0-9.]+)[:\s-]+(.+)$
• Fix: All valid task IDs contain at least one digit (1.1, T1, T15); "Summary" is all-alpha → constrain task ID to require digit
• No existing unit tests for parse_task_content() — tests must be added with fix
• task_status_hook.py may also be affected

Multi-session build state lost (P2)

RT-ICA: APPROVED — 0 blockers

Key discoveries (groomer agent a8c5cfe):

• STATE.md artifact pattern already defined in plugin-creator SKILL.md (lines 41-83) — foundation exists
• Solution: Extend STATE.md with phase completion checklist + commit SHA per phase
• Add agent-results.json tracking task ID → consumed status
• Addresses all 4 related backlog items (intra-phase parallelism, result deduplication, batch fixes, plan divergence)
• No code changes needed — documentation/artifact format update only

SAM: Replace validate-task-file.sh (P2)

RT-ICA: APPROVED — soft blockers

Key discoveries (groomer agent ad62001):

• validate-task-file.sh (170 lines) validates old markdown tasks-refactor-*.md format
• task_format.py exists in python3-development plugin (cross-plugin dependency)
• TASK_FILE_FORMAT.md at .claude/docs/TASK_FILE_FORMAT.md documents target YAML frontmatter format
• Critical: Script not found invoked anywhere — may be unused; verify before implementing replacement
• Cross-plugin dependency decision needed for task_format.py

plugin-validator pre-commit output is too noisy (P1)

RT-ICA: APPROVED — solution approach needs decision

Key discoveries (groomer agent a5b16d6):

• --filter and --filter-type flags already exist in plugin_validator.py but NOT used in pre-commit hook
• commands/development/templates/ directory does NOT exist — FM003 complaint may be stale/resolved
• 4 solution options identified:
  • A: Git staging integration (only validate staged files)
  • B: Severity-level filtering via CLI flag
  • C: Project-level config file for exclusions
  • D: --exclude-codes CLI flag in hook config
• Lowest-effort quick win: Add --show-progress to hook config and use existing --filter-type flag

RT-ICA Results

BLOCKED Items

None — all active items are APPROVED. Three Ideas are blocked by environment constraints:

• Carbonyl Browser Integration: Needs TTY + network
• Validate is-fast: DNS blocked in restricted environment
• Validate agent-browser: Playwright browser download blocked

APPROVED Items

P1 (12 active items):

• plugin-validator UX gaps: 8 conditions verified
• Validate orchestrator-discipline: 9 conditions verified, 1 missing (test methodology — DERIVABLE)
• SAM Extension — Integrate ARL General Theory: 7 conditions verified
• ARL Skill Development: 7 conditions verified
• Meta-Process Capture — Expert Panel: 7 conditions verified
• SAM: Error Recovery / Rollback Procedures: 4 conditions, research-first
• SAM: Human Escalation Criteria: 4 conditions, research-first
• SAM: Timeout/Stall Detection: 4 conditions, research-first
• SAM: Artifact Schema Validation: 4 conditions, research-first
• SAM: Scope Creep Detection: 4 conditions, research-first
• Extract claude-plugin-lint to PyPI: 5 conditions, missing packaging infrastructure
• plugin-validator pre-commit noise: 8 conditions verified, solution approach pending

P2 (16 items): All APPROVED; 8 with research-first dependencies, 8 immediately actionable

Cross-Item Findings

Shared Dependencies

1. plugin_validator.py — Central dependency for 4 backlog items:

   • plugin-validator UX gaps (direct implementation target)
   • plugin-validator pre-commit noise (configuration target)
   • Add PR003/PR004 test coverage (test target)
   • Configurable Token Thresholds (extension target)
   • Caution: Coordinate changes; avoid simultaneous edits

2. ARL reference files (plugins/plugin-creator/skills/arl/references/):

   • SAM Extension — uses synthesis-general-theory.md
   • ARL Skill Development — uses synthesis-arl-applicable.md
   • Meta-Process Capture — uses ARL-agent-instructions.md, qa-expert-panel.md
   • Path correction: Backlog lists skills/assessor/references/ARL/ — actual path is skills/arl/references/

3. bitflight-devops/stateless-agent-methodology repo — Target for all SAM gap items:

   • All 11 SAM gap items (P1 and P2) require cross-repo access
   • Must clone or checkout repo before any SAM implementation work
   • Confirmed by user: SAM has been moved to its own repo

4. /plugin-dev:create-plugin workflow — Source of 4 related P2 items:

   • Multi-session build state, intra-phase parallelism, result deduplication, plan divergence
   • Multi-session state item is the root fix; other 3 are corollaries
   • Address together in one session rather than individually

Suggested Groupings

Research Gaps

Items requiring external research before implementation (all SAM-related):

• GSD rollback/error handling patterns
• BMAD-METHOD human checkpoint patterns
• Temporal/Prefect/Airflow timeout and heartbeat patterns
• AutoGPT task failure handling
• OpenTelemetry for LLM workflow observability

Recommended methodology: /research-and-compare <framework> (in stateless-agent-methodology repo) per backlog Format Guide.

Pre-Existing Issues Found

1. Backlog path error: All ARL-related items reference plugins/plugin-creator/skills/assessor/references/ARL/ — actual path is plugins/plugin-creator/skills/arl/references/. Backlog should be updated.

2. orchestrator-discipline plugin broken: Plugin currently fails claude plugin validate due to invalid rules field. This is a pre-existing functional defect — the plugin is not enforcing delegation discipline as intended.

3. SKILL.md missing user-invocable: true: orchestrator-discipline skill activation is broken until T3 of plan/tasks-4-validate-orchestrator-discipline.md is executed.

4. commands/development/templates/ FM003 complaint: Directory does not exist — pre-commit noise complaint may be partially stale.

5. validate-task-file.sh not integrated: Script exists but is not invoked anywhere — unclear if it serves any active purpose.

Top Three Selected Items

Selection Criteria

1. Priority (P1 > P2)
2. Zero external blockers (fully actionable today)
3. Self-contained in this repository (no cross-repo work)
4. Immediate impact on tooling or security posture
5. Executable by sub-agent with information already gathered

#1: plugin-validator UX and coverage gaps (P1)

Rationale: Highest-impact immediately actionable item. 4 independent sub-issues, all with specific file paths, line numbers, and acceptance criteria. Fixes improve daily developer workflow (false positives suppressed, hooks validated, accurate reporting). No blockers. Plan exists. 19 tests available.

Dependencies: None Prerequisites: None Complexity: Medium — Python code changes to plugin_validator.py

Sub-agent: general-purpose Why: Python implementation task requiring reading source code, making targeted edits to plugin_validator.py, and running tests. General-purpose agent has full tool access for code reading, editing, and test execution.

#2: Validate and verify orchestrator-discipline plugin (P1)

Rationale: Fixes a broken P1 security/enforcement plugin. Plugin currently fails claude plugin validate — it is not functional as intended. Comprehensive plan pre-exists with all architectural decisions made, file paths specified, 5 hook test commands documented. T1/T2/T3 parallelizable; T4 convergence. One focused session can complete all 4 tasks.

Dependencies: None Prerequisites: plan/tasks-4-validate-orchestrator-discipline.md (exists, 4 tasks ready) Complexity: Medium — mix of plugin.json JSON editing, JavaScript hook modification, and SKILL.md frontmatter update

Sub-agent: general-purpose Why: Multi-file task (plugin.json, .cjs hook, SKILL.md, rules/CLAUDE.md) requiring careful coordination. General-purpose agent can follow the plan file and execute all 4 tasks.

#3: Add PR003/PR004 test coverage (P2)

Rationale: Smallest actionable item with zero blockers. Closes a concrete test coverage gap identified in code review. Existing PR001/PR002 test classes serve as copy-paste templates. Severity levels (info vs warning) already verified by groomer agent. No architectural decisions needed — pure test implementation.

Dependencies: None Prerequisites: None (all patterns discovered by groomer agent) Complexity: Small — Python test writing following established patterns

Sub-agent: general-purpose Why: Pure Python test implementation with clear patterns. Groomer agent provided sample test class structures ready to adapt.

Assignment Plan

Item #1: plugin-validator UX and coverage gaps

Agent: general-purpose Instructions to sub-agent:

• Work in plugins/plugin-creator/scripts/plugin_validator.py
• Follow order: Sub-issue 4 (remove lines 904-911 dead code + error strings at 759-760, 771-772) → Sub-issue 1 (fix reporter at lines 2928-2931 to count unique files, not validators) → Sub-issue 3 (note: FileType.HOOK_CONFIG already exists at lines 318-329; add HOOK_SCRIPT, add HookValidator class, add HK001+ codes) → Sub-issue 2 (add file_type check to DescriptionValidator, suppress SK005 for COMMAND files)
• Run uv run pytest plugins/plugin-creator/tests/ -v after each sub-issue
• Reference QA report at plugins/plugin-creator/planning/plugin-validator-qa-report.md for context on 15 pre-existing failures
• Reference plan at plan/tasks-2-validator-ux-coverage.md

Item #2: Validate orchestrator-discipline plugin

Agent: general-purpose Instructions to sub-agent:

• Read plan/tasks-4-validate-orchestrator-discipline.md in full before starting
• Execute T1 (fix plugin.json rules field), T2 (add directory detection to hook), T3 (add user-invocable: true to SKILL.md) in parallel if possible, then T4 (convergence validation)
• Run claude plugin validate plugins/orchestrator-discipline/ after T1
• Run the 5 hook test commands specified in plan after T2
• Run uv run prek run --files plugins/orchestrator-discipline/ after T3
• Document any pre-existing issues found in T4 output per plan line 800-801

Item #3: Add PR003/PR004 test coverage

Agent: general-purpose Instructions to sub-agent:

• Target file: plugins/plugin-creator/tests/test_plugin_registration_validator.py
• Read existing PR001 (lines 267-328) and PR002 (lines 470-543) test classes for patterns
• PR003 = info severity → check result.info list; emitted when metadata fields (repository, homepage, author) absent
• PR004 = warning severity → check result.warnings list; emitted when plugin.json repository differs from git remote URL
• Create TestMissingMetadata class for PR003 tests
• Create TestRepositoryMismatch class for PR004 tests — may need monkeypatch for git metadata
• Run uv run pytest plugins/plugin-creator/tests/test_plugin_registration_validator.py -v

Backlog Corrections Needed

1. Update path for all ARL-related items: skills/assessor/references/ARL/ → skills/arl/references/
2. Promote "Configurable Token Thresholds" from Idea to P2 (zero blockers, small scope)
3. Add note to "SAM: Replace validate-task-file.sh" — verify script is actually used before implementing replacement
4. Mark "Validate is-fast", "Validate agent-browser", "Validate carbonyl" as BLOCKED pending environment access