Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
These are local-only planning documents, now covered by .gitignore. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
cb045ee to
96aabad
Compare
Handle fumadocs index slug stripping in mdx route, remove dead getLLMTextByPaths, fix SKILL.md Content-Type to text/markdown, and standardize CONTENT constant naming. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Pin all GitHub Action third-party refs to full commit SHAs
- Move ${{ }} expressions to env: vars to prevent shell injection
- Create shared LLM_FILES config (single source of truth for manifest,
sitemap, and token counter)
- Re-run token counter after Action content updates
- Return 410 Gone from deprecated /llms-full.txt endpoint
- Add sitemap reference to robots.txt
- Standardize all routes to module-level const CONTENT
- Fix SKILL.md sed pattern in Action after CONTENT rename
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…-script
- Fix Solana/Sui progressive disclosure URLs (add pull-integration/ segment)
- Remove per-request Generated on timestamp from all route handlers
(contradicts revalidate=false, latent CDN caching bug)
- Move sitemap lastModified to module-level BUILD_DATE constant
- Move remaining ${{ }} expressions in github-script to env/process.env
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
@codex review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f10cacae55
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
|
|
||
| - name: Read static headers from affected product files | ||
| - name: Read content from affected product files | ||
| if: steps.changed-files.outputs.significant == 'true' && steps.changed-files.outputs.affected_products != '' |
There was a problem hiding this comment.
Run SKILL review even when no product route is mapped
This gate makes the entire update pipeline skip whenever affected_products is empty, even if include_skill was set to true. A PR that only changes key docs outside the hardcoded product mapping (for example content/docs/express-relay/index.mdx or content/docs/metrics/index.mdx) will be marked significant but never evaluate/update SKILL.md, which defeats the new include_skill logic and leaves the playbook stale for those docs changes.
Useful? React with 👍 / 👎.
| if [ "$INCLUDE_SKILL" = "true" ]; then | ||
| SKILL_FILE="$ROUTE_DIR/SKILL.md/route.ts" | ||
| if [ -f "$SKILL_FILE" ]; then | ||
| sed -n '/const CONTENT = `/,/^`;/{/const CONTENT = `/d;/^`;/d;p}' "$SKILL_FILE" > "/tmp/header-SKILL.md.txt" |
There was a problem hiding this comment.
Preserve SKILL frontmatter when extracting CONTENT
This sed extraction deletes the entire const CONTENT = ... line, but SKILL.md starts its markdown payload on that same line (---). As a result, the opening frontmatter delimiter is dropped before sending content to the model, and any rewritten SKILL.md can lose valid YAML frontmatter structure after update.
Useful? React with 👍 / 👎.
Fix two bugs in update-llms-txt workflow: 1. sed extraction was deleting the entire `const CONTENT` marker line, silently dropping the first line of content (frontmatter `---` in SKILL.md, `# Heading` in product files). Now strips only the prefix. 2. Pipeline steps were gated on `affected_products != ''`, skipping SKILL.md review when doc changes don't map to a product route. Now also runs when `include_skill == 'true'`. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Summary
Restructures the LLM-consumable documentation system from a monolithic llms full.txt approach to a Tiered + Protocol Hybrid architecture. AI agents can now progressively load context -- starting with a lightweight routing index (~500 tokens), drilling into curated product guides (~2k tokens each), and optionally fetching individual .mdx pages for full detail. This replaces the old architecture where a single monolithic file dumped all documentation at once, wasting token budget on irrelevant content. A new machine-readable manifest, dynamic sitemap, token counting toolchain, and hardened GitHub Action CI round out the system.
Rationale
How has this been tested?