Skill Align

Lattice Sync

Core responsibility: Keep every public-facing document in sync with the actual skill inventory. The skills directory is the source of truth — all documents derive from it.

Input: None required. Reads the live state of skills/ and all documentation on every run. Optionally: a specific skill name or tier to focus the audit.

Output:

• A findings report listing every gap as [GAP], [STALE], or [WRONG] with file and description
• All found gaps fixed in-place across: README.md, docs/how-it-works.md, docs/configuration.md, docs/practical-guide.md, PROJECT.md, CLAUDE.md, .github/ISSUE_TEMPLATE/bug_report.yml
• A final clean confirmation: "No gaps found" or a list of what was changed

How to verify this skill did its job:

grep -rn "requirements atom\|framework:requirements\b" \
  docs/ README.md PROJECT.md CLAUDE.md .github/ skills/ \
  --include="*.md" --include="*.yml"

Any result from this grep means the sync is incomplete. A clean run returns no output.

Also verify: every skill in skills/ appears in the bug_report.yml dropdown and the how-it-works.md inventory tables.

When to run

• After creating any new atom, molecule, or refiner
• After renaming or removing a skill
• After changing a skill's config key, consumed-by relationships, or output path
• When a user suspects docs are out of date

Phase 1 — Build the live inventory

Read the skills directory tree. STOP: do NOT assume you already know the inventory — always read from files.

find skills/ -name "SKILL.md" | sort

For each SKILL.md found, extract:

• name: field (frontmatter)
• tier: inferred from path (atoms/, molecules/, refiners/)
• Config key: grep for paths\.\w+ in the file
• Consumes: grep for framework: references (molecules only)
• Produces: grep for .lattice/standards/ path (refiners only)
• Output subfolder: grep for .lattice/\w+/ paths (molecules only)

Build a structured inventory:

atoms:    [name, config_key, has_refiner, has_defaults_md]
molecules: [name, composes[], output_subfolder]
refiners:  [name, config_key, produces_path, consumed_by_atom]

Read references/audit-checklist.md for the complete per-document audit rules before starting Phase 2.

Phase 2 — Audit each document

Work through every document in this order. For each, apply the checks in references/audit-checklist.md. STOP: a document with zero findings gets no entry in the report — do not print a heading or a "no gaps" line for it. Log every finding as:

[GAP]    file:line — description of what is missing or wrong
[STALE]  file:line — description of what refers to something that no longer exists
[WRONG]  file:line — description of a relationship (consumed-by, produces) that is incorrect

Documents to audit:

1. docs/how-it-works.md
2. docs/configuration.md
3. docs/practical-guide.md
4. README.md
5. PROJECT.md — Skill Conventions, Known subfolders list, and Repository Structure counts all live here
6. .github/ISSUE_TEMPLATE/bug_report.yml
7. .github/ISSUE_TEMPLATE/skill_request.yml (generic — only check if skill-specific examples are present)
8. .github/ISSUE_TEMPLATE/documentation.yml (generic — only check if skill names appear)
9. knowledge-base/requirement-forge-requirements.md (if exists — check it matches current skill names)

Present a consolidated findings report before making any changes:

## Lattice Sync — Findings

### docs/how-it-works.md
[GAP] Atoms table missing: requirement-quality
[STALE] Refiners in Depth: requirement-forge-refiner consumed-by says "molecule" but should say "requirement-quality atom"

### .github/ISSUE_TEMPLATE/bug_report.yml
[GAP] Skill dropdown missing: requirement-quality (atom)

Total: N gaps, M stale references, P wrong relationships

Ask: "Ready to apply all fixes? Or are there any findings you want to skip?"

Phase 3 — Fix

Apply every agreed fix. For each document, make all changes in a single edit pass. STOP: do not make multiple passes over the same file.

After all fixes are applied, run one final verification grep to confirm no old references remain:

# Check for stale atom names or wrong consumed-by text
grep -rn "<old-name>\|consumed.*molecule\|consumed.*wrong" \
  docs/ README.md PROJECT.md CLAUDE.md .github/ --include="*.md" --include="*.yml"

If the grep returns results, fix them before declaring done.

Phase 4 — Deploy (optional)

If the user wants to push the updated skills to their AI tool's skills directory:

./tools/install.sh /path/to/your/skills/folder

Ask the user for the target path if not provided. Default for Claude Code: ~/.claude/skills/.