Internal guide for maintaining the doc-based skills in this repo. End users do not need to read this — the skills work as-is once installed.
The following skills under skills/ track upstream documentation in tidev/titanium-docs:
| Skill | Upstream subtree |
|---|---|
ti-api |
docs/api/ |
ti-guides |
docs/guide/Titanium_SDK/Titanium_SDK_Guide |
ti-howtos |
docs/guide/Titanium_SDK/Titanium_SDK_How-tos |
alloy-guides |
docs/guide/Alloy_Framework/Alloy_Guide |
alloy-howtos |
docs/guide/Alloy_Framework/Alloy_How-tos |
When upstream docs change, the reference files inside each skill should be re-aligned. This is the workflow.
Audits run on demand, not on a schedule. The trigger is upstream change:
- A new release or notable commit lands in
tidev/titanium-docs. - You notice drift between a reference and current SDK behavior.
- A user reports that a skill's content is stale or incorrect.
Audit only the skill whose upstream subtree changed. The mapping is in the table above. If the upstream change is under docs/guide/Alloy_Framework/Alloy_Guide/, only audit alloy-guides — not the other three. Running unnecessary audits wastes time and risks introducing churn from unrelated diffs.
If you're not sure which skill maps to an upstream change, run git log -- <path> inside .titanium-docs/ and cross-reference the table above.
A maintenance-only skill ships at .claude/skills/ti-skill-audit/. It is auto-discovered by Claude Code when working inside this repo and is not part of the public skills/ set.
The audit is a 7-phase process with a hard-stop approval gate. See .claude/skills/ti-skill-audit/references/audit-workflow.md for the full spec.
The audit compares reference files against a local clone of tidev/titanium-docs at .titanium-docs/ in the repo root. The folder is gitignored.
git clone --depth 1 https://github.com/tidev/titanium-docs.git .titanium-docsIn Claude Code, invoke:
/ti-skill-audit alloy-howtos
The auditor will:
- Verify
.titanium-docs/exists; if missing, prompt you to clone. - Optionally
git pull --ff-onlyto refresh the cache. - Note the upstream commit hash for the audit report.
- Run coverage analysis (Phase 1) and gap identification (Phase 2).
- Generate a consolidated report (Phase 3).
- Stop and wait for your explicit approval (Phase 4).
- Apply the approved updates and verify (Phases 5–6).
Before each audit:
cd .titanium-docs && git pull --ff-only && cd ..One commit per audited skill, e.g.:
audit(alloy-howtos): align refs with titanium-docs@a3f2b1c
Mention the upstream commit hash from Phase 0 in the commit body or PR description.
Each reference file may contain a ## Community-Discovered Patterns (H2) section. These sections capture verified real-world patterns that are intentionally not in the official docs. The auditor will never delete them during a sync.
If an upstream release adds official coverage for a pattern, the auditor will move that content into the main section and cite the source. If SDK behavior changed, it will update the pattern. An outdated workaround is just as bad as a hallucinated one. The H2 protects the section, but doesn't freeze it.
Some reference files are sourced from elsewhere and should not be compared against tidev/titanium-docs at all. They carry an HTML comment near the top:
<!-- AUDIT-SKIP: ... reason ... -->The auditor reads that marker in Phase 0 and excludes the file from every later phase. Current example: skills/alloy-guides/references/PURGETSS.md, which is maintained against the PurgeTSS toolkit's own docs at purgetss.com, not against the Alloy guide upstream. To update one of these files, edit it manually.
The ti-skill-audit skill is Claude-Code-specific (it lives under .claude/skills/). Maintainers using other agents can:
- Read
.claude/skills/ti-skill-audit/references/audit-workflow.mdas a manual checklist. - Apply the same workflow by hand, comparing files in
.titanium-docs/against the correspondingreferences/folder.
The methodology stays the same regardless of agent: coverage analysis, gap identification, hard-stop approval, preserve community patterns, respect AUDIT-SKIP markers.
Every doc-based skill follows:
skills/<skill-name>/
SKILL.md # Entry point (~200–500 lines)
references/
TOPIC_ONE.md # Deep reference (~200–800 lines each)
TOPIC_TWO.md
...
Per AGENTS.md and the agentskills.io specification:
- Only
nameanddescriptionare guaranteed to be portable across agents. descriptionmust start withUse when…and describe trigger conditions, not workflow.- Total frontmatter ≤ 1024 characters.
- Skills are agent-neutral — no Claude-specific fields like
argument-hintorallowed-toolsinskills/.
The .claude/skills/ti-skill-audit/ skill is exempt from cross-agent rules because it lives under .claude/skills/ (Claude Code-only by location).
- Agentskills.io spec: https://agentskills.io/specification
- TiDev (current Titanium maintainer): https://github.com/tidev
- Titanium docs source: https://github.com/tidev/titanium-docs
- Titanium SDK: https://github.com/tidev/titanium-sdk