Cleaning and optimizing skills

[OnoSendai13]

ecc-skill-stocktake-proposal.md

Skill Stocktake — Cleanup & Consolidation Proposal

I ran a full stocktake audit across all ~88 skills in ECC. I'm proposing the following changes to the skill catalog. These are based on content overlap analysis, freshness checks, and structural evaluation — not arbitrary pruning.

---

Changes Applied in This PR

1. Retire continuous-learning (v1)

Status: Fully superseded by continuous-learning-v2

• v1 explicitly lists "Potential v2 Enhancements" that are already implemented in v2 (confidence scoring, project isolation, promotion pipeline)
• The v1 comparison table shows v2 is strictly superior (hooks-based vs Stop hook, instincts vs full skills)
• Keeping both creates confusion about which to use
• Action: Delete skills/continuous-learning/

2. Retire project-guidelines-example

Status: Not a reusable skill — it's a project-specific template

• Explicitly labeled as an example for a specific project (Zenith)
• Contains hardcoded project-specific architecture, file paths, deployment details
• Action: Delete skills/project-guidelines-example/ (or move to docs/templates/)

3. Retire coding-standards — merge into rules/common/coding-style.md

Status: Massive overlap with other skills

• Immutability, error handling, async patterns, React component structure → already in frontend-patterns with more depth
• API response format, rate limiting, authentication → already in api-design and backend-patterns
• Universal principles (KISS, DRY, YAGNI) → better suited as project rules, not a skill
• Action: Delete skills/coding-standards/ and add universal principles (KISS, DRY, YAGNI, code smells) to rules/common/coding-style.md

File changes summary

 rules/common/coding-style.md              | +28 (added KISS, DRY, YAGNI, code smells)
 skills/coding-standards/SKILL.md          | -530 (deleted)
 skills/continuous-learning/SKILL.md       | -100 (deleted)
 skills/continuous-learning/config.json    |  -20 (deleted)
 skills/continuous-learning/evaluate-session.sh | -N (deleted)
 skills/project-guidelines-example/SKILL.md |  -N (deleted)

Net: +28 lines of improved rules, ~1085 lines of redundant skill content removed.

---

Recommendations for Future PRs (Not Included Here)

I'm flagging these as "consider for later" since they require more judgment calls. I'm happy to open separate PRs for any of these:

Skills to Improve (actionable items)

Skill	Issue	Suggested fix
django-verification	12 phases too heavy; pip-audit AND safety check redundant; assumes PostgreSQL without noting it	Consolidate security section, note DB-specific assumptions, trim to 8-10 phases
backend-patterns	Response envelope section duplicates api-design; Supabase-specific examples too narrow for generic title	Remove envelope section, generalize examples, or rename to reflect Supabase/Next.js focus
iterative-retrieval	Pseudocode examples not directly usable; external reference (Jan 2025) may be stale	Add concrete language-specific implementations, update/remove dated external references
mcp-server-patterns	68 lines, mostly defers to external docs	Add at least one complete working server example (tool registration, resource handler, stdio transport)
playwright-cli	279-line CLI catalog, not a skill	Transform into usage patterns (login flow, form validation, SPA nav) with decision frameworks
plankton-code-quality	80% about one external tool's architecture	Simplify to focus on integration patterns, remove opinionated package-manager blocking
tdd-workflow	Very TypeScript/Next.js-specific but not labeled as such	Make scope explicit, or split into generic template + TS-specific examples
verification-loop	High structural overlap with springboot-verification	Ensure springboot-verification explicitly extends this rather than duplicating the report format

Remotion rule consolidation (in-skill, not cross-skill)

remotion-best-practices uses a hub+37-rules pattern. 6 thin rule files could be merged into broader rules:

• animations.md (27 lines) → timing.md
• ffmpeg.md (34 lines) → trimming.md + videos.md
• get-audio-duration.md (58 lines) → audio.md
• get-video-dimensions.md (68 lines) → videos.md
• get-video-duration.md (61 lines) → videos.md
• import-srt-captions.md (69 lines) → display-captions.md

This would reduce the Remotion rule count from 37 to ~31 with zero loss of content.

---

Methodology

The audit followed the skill-stocktake skill's checklist:

1. Inventory scan — enumerated all skills with mtime and usage frequency (7d/30d)
2. Content overlap analysis — checked each skill against other skills and against rules/common/ files
3. Freshness check — verified technical references against current versions
4. Usage frequency — considered (all skills show 0/30d, which is expected for domain-specific skills)
5. Holistic judgment — verdict based on actionability, scope fit, uniqueness, and currency

All three batches were evaluated in parallel by sub-agents reading each skill file independently.

[affaan-m]

Shipped the first safe stocktake slice on main in 0f4f95b.\n\nWhat changed:\n- moved the non-reusable \ skill out of the live catalog into \n- removed it from install manifests\n- updated configure-ecc/docs references to point at the docs example instead of a live skill\n- synced counts to 158 skills\n\nI am leaving the broader issue open because the larger overlap/consolidation work still remains.

[affaan-m]

Shipped the first safe stocktake slice on main in 0f4f95b.

What changed:

• moved the non-reusable project-guidelines-example skill out of the live catalog into docs/examples/project-guidelines-template.md
• removed it from install manifests
• updated configure-ecc and docs references to point at the docs example instead of a live skill
• synced counts to 158 skills

I am leaving the broader issue open because the larger overlap/consolidation work still remains.

[affaan-m]

Shipped another consolidation slice on main in 92fbc52. Tightened product-lens so it now explicitly owns product diagnosis / brief validation, while the new product-capability lane owns implementation-ready capability plans and SRS-style constraints. This removes a fresh overlap instead of letting both skills claim the same planning surface.

[affaan-m]

Another safe cleanup slice is now on main in 4b24d57.

Shipped in this pass:

• removed the deleted project-guidelines-example skill from the exported agent.yaml inventory
• repointed contributor docs to docs/examples/project-guidelines-template.md
• marked continuous-learning v1 as the supported legacy path and pointed new installs toward continuous-learning-v2

This keeps the repo honest about what is actually shipped without deleting the still-referenced coding-standards or continuous-learning surfaces prematurely.

[affaan-m]

Another cleanup slice is now on main in b3a56f8.

This pass removed the last orphaned localized copies of the deleted project-guidelines-example skill from docs/ko-KR and docs/zh-CN. The project-specific template now lives only in docs/examples/project-guidelines-template.md, which matches the current repo surface.

Validation stayed green with node scripts/ci/validate-skills.js and node scripts/ci/check-unicode-safety.js.

[affaan-m]

Shipped another safe slice directly to in .\n\nThis pass keeps as a real shared surface, but narrows it into the baseline cross-project conventions layer instead of letting it compete with narrower implementation skills.\n\nWhat changed:\n- rewrote the skill description to focus on naming, readability, immutability, and code-quality review\n- added explicit scope boundaries so React/UI architecture defers to \n- pointed backend/API structure to and \n- kept as the lightest reusable rule layer\n\nValidation stayed green with Validated 180 skill directories and Unicode safety check passed..

[affaan-m]

Shipped another safe #1213 slice directly to main in c547823.

This pass keeps coding-standards as a real shared surface, but narrows it into the baseline cross-project conventions layer instead of letting it compete with narrower implementation skills.

What changed:

• rewrote the skill description to focus on naming, readability, immutability, and code-quality review
• added explicit scope boundaries so React/UI architecture defers to frontend-patterns
• pointed backend/API structure to backend-patterns and api-design
• kept rules/common/coding-style.md as the lightest reusable rule layer

Validation stayed green with node scripts/ci/validate-skills.js and node scripts/ci/check-unicode-safety.js.

[affaan-m]

Landed another overlap-cleanup slice on main in 572c7a8.

Shipped:

• docs/capability-surface-selection.md
• linked from skills/mcp-server-patterns/SKILL.md

This makes the repo boundary explicit:

• rules/ = deterministic, always-on constraints
• skills/ = on-demand workflows and token-expensive playbooks
• MCP = structured interactive capabilities that justify a long-lived server surface
• local CLI/scripts = simple deterministic one-shot execution
• direct API calls = narrow remote steps that do not yet justify MCP

That is also the policy for external ideas going forward: copy the underlying idea, then repackage it into the narrowest ECC-native surface instead of standardizing another plugin/runtime by default.

[affaan-m]

Shipped another #1213 slice directly to main.

Now live in 6c1c58b:

• added docs/skill-adaptation-policy.md
• linked that policy from CONTRIBUTING.md

This codifies the repo rule we keep enforcing manually:

• copy the underlying idea, not the external product identity
• rename when ECC materially changes or expands the surface
• prefer ECC-native rules / skills / scripts / MCPs over new default third-party dependencies
• do not ship a skill whose main value is telling users to install an unvetted package

This is the policy version of the #370 lesson: if a contribution only makes sense as another product's branded wrapper, it either needs to be repackaged into ECC-native form or not shipped.

[affaan-m]

Shipped another safe #1213 slice directly to main in 1b3ccb8.

This pass makes the install-facing catalog stop presenting continuous-learning v1 as equivalent to continuous-learning-v2.

What changed:

• marked skill:continuous-learning as a legacy v1 Stop-hook surface in manifests/install-components.json
• updated the workflow-quality module description to call out the legacy v1 path explicitly
• updated skills/configure-ecc/SKILL.md so the selector table tells new installs to prefer continuous-learning-v2
• updated the top-level README.md directory/catalog text so continuous-learning is labeled as legacy and /skill-create --instincts points at continuous-learning-v2
• added a regression test in tests/lib/install-manifests.test.js to keep the legacy labeling from drifting back

Validation stayed green with:

• node tests/lib/install-manifests.test.js
• node tests/lib/selective-install.test.js
• node scripts/ci/validate-skills.js
• node scripts/ci/check-unicode-safety.js

I am leaving #1213 open because the broader overlap/consolidation work still remains, but the install/catalog surface is now honest about v1 vs v2 learning.