ci: add daily upstream sync + custom skills lint

[royosherove]

Summary

Automated daily sync from upstream Kiro Powers + CI validation for custom skills consistency.

Changes

Sync Workflow (sync-upstream.yml)

• Schedule: Daily at 08:00 UTC (11:00 Israel time)
• Action: Clone upstream → copy new/updated skills → create PR if changes
• Custom skills preserved: 8 skills (module-organizer, refactoring, unit-testing, claude-agent-sdk, cross-agent-test, lambda-durable, mcporter, playwright-cli)
• Safety: Clones from main (no release tags), filters SKILL.md only, SHA-pinned actions
• Notifications: Creates PR (manual review before merge)

Lint Workflow (lint.yml)

• Trigger: Every push/PR to any branch
• Check: CUSTOM_SKILLS array (workflow) ↔ README table (sync requirement)
• Failure: Blocks merge if lists drift
• Speed: ~5-10s validation

Documentation (README.md)

• Expanded custom skills table (8 items with descriptions)
• Added sentinel comments for markdown anchoring
• Updated sync instructions with automation details
• Clarified prerequisites (GitHub Actions approval setting)

Security

• ✅ All actions pinned to commit SHA + version comment
• ✅ Grep patterns anchored to prevent false matches
• ✅ Markdown sentinels for robust extraction

Testing

• 10 review cycles (supply-chain hardening, edge case handling, consistency validation)
• All checks passed (lint + architect review)

Ready to merge!

[github-actions]

Claude finished @royosherove's task in 1m 24s —— View job

---

PR Review

• Read repo context
• Review sync-upstream.yml
• Review lint.yml
• Review README.md changes
• Post final review

---

Overall this is well-structured and clearly thought through. A few issues worth addressing before merge.

---

sync-upstream.yml

Bug — labels will be treated as a single string (line 130)

labels: 'chore, automated'

peter-evans/create-pull-request accepts labels as newline-separated values. This will create one label named "chore, automated" rather than two labels. Fix:

labels: |
  chore
  automated

Fix this →

---

Supply chain risk — upstream cloned unpinned from main (line 33)

git clone --depth 1 https://github.com/kirodotdev/powers.git /tmp/powers

This is acknowledged in the PR description ("no release tags"). It means a compromised upstream main branch would directly sync malicious content into a PR here. Since you cannot pin to a commit SHA on an external repo you don't control, the mitigation is the manual PR review step — which is already in place. Worth calling out explicitly in the workflow comments so future maintainers don't remove the PR creation step thinking it's just overhead.

---

$skill_name used in rm -rf without validation (lines 56–67)

skill_name=$(basename "$skill_dir")
...
rm -rf "$skill_name"

basename protects against .. traversal, but a skill named -rf or containing special characters could still cause unexpected behavior. A simple guard would help:

# Validate: skill name must be alphanumeric + hyphens only
if [[ ! "$skill_name" =~ ^[a-zA-Z0-9_-]+$ ]]; then
  echo "⚠️  Skipping invalid skill name: $skill_name"
  continue
fi

Fix this →

---

lint.yml

Fragility — single-line array assumption (line 19)

WORKFLOW_SKILLS=$(grep "^[[:space:]]*CUSTOM_SKILLS=(" .github/workflows/sync-upstream.yml \
  | sed 's/.*CUSTOM_SKILLS=(\([^)]*\)).*/\1/' \
  | tr ' ' '\n' | sed 's/"//g' | sort)

This only works when CUSTOM_SKILLS=(...) is declared on a single line. If someone reformats the array to multi-line (e.g., for readability when more skills are added), $WORKFLOW_SKILLS becomes empty and the check silently passes a false positive or fails. Consider a comment warning about this constraint:

# NOTE: CUSTOM_SKILLS must remain on a single line in sync-upstream.yml
# Multi-line format will break this extraction.

Or switch to a more robust extraction approach.

Fix this →

---

Missing permissions block

The lint job performs only reads, but has no explicit permissions declaration. Best practice for least-privilege:

jobs:
  verify-custom-skills-sync:
    runs-on: ubuntu-latest
    permissions:
      contents: read

---

README.md

Changes look clean. The sentinel comments are well-placed and the note inside the sentinel block won't be incorrectly parsed by the lint grep (which anchors to | \...``).

---

Summary

Severity	Issue	File
Bug	labels creates single label instead of two	sync-upstream.yml:130
Medium	No $skill_name validation before rm -rf	sync-upstream.yml:67
Low	Lint breaks if CUSTOM_SKILLS is reformatted to multi-line	lint.yml:19
Low	Missing permissions: contents: read	lint.yml
Info	Upstream clone is unpinned (acknowledged, PR review mitigates)	sync-upstream.yml:33

The labels bug is the only one that will silently misbehave at runtime. The rest are defensive hardening. Good work overall — the structure, SHA-pinning, and concurrency handling are solid.