All changes made to the Master Starter Kit and supporting primitives on April 12, 2026. 25 items across 5 batches.
| Item | File(s) | Before | After |
|---|---|---|---|
| SETUP.md skill count | Project Starter/SETUP.md |
215 curated skills | 297 curated skills |
| SETUP.md hook count | Project Starter/SETUP.md |
59 curated hooks | 46 curated hooks |
| SETUP.md workflow count | Project Starter/SETUP.md |
38 curated workflows | 40 curated workflows |
| SETUP.md command count | Project Starter/SETUP.md |
13 slash commands | 46 slash commands |
| PROCESS.md cascade ordering | PROCESS.md |
Arsenal-autonomy listed as primary cascade mechanism (#1); sync-arsenal.sh listed #3 | sync-arsenal.sh is primary (#1); arsenal-autonomy is #3 with explicit note about SessionStart unreliability |
| PROCESS.md cascade checklists | PROCESS.md |
No explicit cascade steps in the 6 process workflows | All 6 workflows (New Item, New Hook, New Workflow, New Plugin, New Agent, Update Existing) now include cascade checklist with sync-arsenal.sh and validate-counts.sh |
| Item | File(s) | Before | After |
|---|---|---|---|
| LICENSE file | Project Starter/LICENSE |
No license file existed | MIT License, Copyright (c) 2026 TBK Labs |
| PS-native hooks documentation | Project Starter/PS-NATIVE-HOOKS.md |
7 hooks in PS had no canonical source in Curated, undocumented gap | New file documents all 7 PS-native hooks: 5 flagged as promotion candidates, 2 documented as intentionally PS-only |
| validate-counts.sh | Project Starter/scripts/validate-counts.sh |
Did not exist. Count claims could drift silently. | 107-line script that counts actual directories in Curated and PS, compares against SETUP.md claims, checks PS/Curated alignment, exits non-zero on mismatch |
| preflight.sh integration | Project Starter/scripts/preflight.sh |
Ran format, type, and test checks only | Now also runs validate-counts.sh as a warning step before final pass/fail |
| Curated-to-PS drift | 49 files across Project Starter/.claude/skills/ |
49 skill files had content differences between Curated and PS copies | All 49 files re-synced from Curated to PS via bulk copy |
| Arsenal-autonomy trigger model | .autonomous/registry.json (both root and PS) |
Loops implied automatic triggering via SessionStart hook (unreliable) | All 4 loops have "trigger": "manual"; orchestrator SKILL.md updated to say "invoked manually via slash command" |
| Item | File(s) | Before | After |
|---|---|---|---|
| TODO cleanup | context-handoff/SKILL.md, whatsapp-cloud-api/SKILL.md |
Contained "TODO: implement if needed" and placeholder TODO comments | Replaced with proper descriptions and documentation |
| Related Skills cross-references | 14 agent-cluster SKILL.md files | 14 overlapping agent skills with no cross-references | All 14 now have "## Related Skills" sections organized into 3 sub-clusters: Agent Architecture (6), AI Infrastructure (3), AI Development (5) |
| Skill type system | All 297 Skills/Curated/*/SKILL.md |
No type classification. Skills had no machine-readable behavior indicator. | Every SKILL.md has type: in YAML frontmatter: 248 workflow, 37 hybrid, 12 reference. All cascaded to PS. |
| SKILL-INDEX.md | Skills/Curated/SKILL-INDEX.md + PS copy |
Only alphabetical directory listing for skill discovery | Use-case-based index with 14 task categories (Code Quality, Architecture, Frontend, Backend, DevOps, Security, Testing, AI/Agents, Content/Marketing, PM, Documentation, Design/UX, Data, Platforms) |
| Item | File(s) | Before | After |
|---|---|---|---|
| Thin skills expanded (Batch 4a: 8) | debug-failing-test, merge-conflict, api-review, benchmark, email-sequence, a11y-debugging, launch-strategy, troubleshooting | Under 100 lines each, minimal content | Expanded to 100-300+ lines with proper workflow steps, examples, and edge cases |
| Hook dependency guards | 19 Hooks/Curated/*/hook.sh files |
Hooks assumed tools like jq, shellcheck, markdownlint were installed; would fail with cryptic errors if missing |
All 19 now have command -v TOOL || exit 0 guards that skip gracefully with "[HOOK SKIP]" message if dependency is missing |
| TTG protocol clarification | ORCHESTRATOR-FULL.md |
TTG protocol said "think then generate" but was ambiguous about when to show reasoning vs. resolve internally | Phase-aware interaction model: Discovery (0-1.9) shows reasoning and waits; Research/Architecture (2-5.9) resolves internally, asks only for irreversible decisions; Generation (6+) resolves entirely internally. "When in doubt, resolve internally." |
| Enforcement gate path-awareness | ENFORCEMENT-GATES.md |
17 gates applied uniformly to all projects regardless of path | New "Path Awareness" section with 17-gate x 6-path applicability table (Express, Lite, Standard, Full, Library/CLI, Migration) using Skip/Advisory/Full/Conditional enforcement levels. Planning vs. Implementation mode distinction. |
| 9 path guides created | 9 new *-PATH-GUIDE.md files in Master-Starter-Kit/ |
PATHS.md listed paths but no dedicated guide for any path | EXPRESS (299 lines), STANDARD (124), FULL (184), LITE (283), MIGRATION (358), LIBRARY-CLI (334), GUIDED (344), ENHANCE (389), REPURPOSE (422). Each specifies phases, gates, skips, and time estimates. |
| Sync verification | Project Starter/scripts/sync-arsenal.sh |
--verify flag existed but did nothing meaningful |
New verify_sync() function (~85 lines) counts skills/hooks, diffs random sample of 5 skills, reports mismatches |
| Manual fallback headers | All 5 scripts in Project Starter/scripts/ |
Scripts had no fallback instructions if they couldn't run | All 5 scripts (sync-arsenal.sh, validate-counts.sh, preflight.sh, setup.sh, cascade-product-spec-rule.sh) now have comment blocks at top explaining how to do each check manually |
| Item | File(s) | Before | After |
|---|---|---|---|
| Scheduled cascade check | Scheduled task: arsenal-cascade-check |
No automated drift detection between Curated and PS | Weekly Sunday 10am task checks for drift, saves report to audits/cascade-check-YYYY-MM-DD.md |
| Thin skills expanded (remaining 23) | 23 additional SKILL.md files across 3 batches | Under 100 lines each | All expanded to 100-300+ lines. Includes: executing-plans, form-cro, investor-materials, telegram, instagram, llm-app-patterns, autonomous-agent-patterns, audio-transcriber, product-inventor, voice-ai-engine, stitch-ui-design, aidesigner-frontend, json-canvas, chief-security-officer-brief, microsoft-docs, pareto-mac, wordpress-penetration-testing, claude-mem-openclaw, qa-only, agent-orchestrator, web-design-review, hugging-face-jobs, paywall-upgrade-cro |