Repository navigation
chore: help agents find hyperframes on npm and context7 - #5937
Conversation
Agents asked to make a video choose from what is on disk. The AGENTS.md and CLAUDE.md that init writes now open with a plain statement of what the project is and that `hyperframes render` replaces a hand-built headless-browser and ffmpeg pipeline, and list `snapshot` beside preview, check, render and docs. A test pins that init writes both files from the template, and the CLI docs now say init creates them. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
npm search ranks on description and keywords. The hyperframes package now says
what it does ("Make videos with code and AI agents: write HTML, render MP4.
Open source.") and lists the words people and agents search for. Core and
studio had empty descriptions; producer and player gain keywords too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The opening paragraph said what HyperFrames is but not how an agent starts. Add one line: install the skills, then ask the agent for a video. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Preview deployment for your docs. Learn more about Mintlify Previews.
💡 Tip: Enable Automations to automatically generate PRs for you. |
Inside a scaffolded project the agent already uses the CLI, so telling it not to build its own capture pipeline there adds nothing. The intro now only says what the folder is. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Edit accuracy: accurate 271 (base branch 271), smooth 161 of thoseThe gate passes. Quarantined, measured but not gated (0) Unstable (2)
|
Internal packages are not searched on npm, and the init test only re-checked the template's own text. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Indexes docs, skills, registry and examples only (no test fixtures), and hands agents five rules from AGENTS.md. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ind-hyperframes # Conflicts: # packages/cli/package.json # packages/cli/src/templates/_shared/AGENTS.md # packages/cli/src/templates/_shared/CLAUDE.md
jrusso1020
left a comment
There was a problem hiding this comment.
The CLI package change is limited to its description and 14 unique keywords; it does not claim an MCP server. The shared AGENTS.md and CLAUDE.md templates remain byte-identical, init copies both, and the added snapshot --at 1,3,5 example matches the command parser.
context7.json validates against the official Context7 schema, including all five rules within the 255-character limit. Its include/exclude settings and agent guidance are consistent with the intended documentation scope. This does not prove that Context7 has reindexed or improved retrieval; account claiming, reindexing, repository topics and the reported model evaluation were not exercised.
No package publish, full init run, full CLI suite or live model evaluation was performed. Prior review was only a Copilot quota notice.
Verdict: COMMENT
Reasoning: No code blocker found in metadata, indexing configuration or the small documentation additions. Approval needs James’s go or a trusted stamper’s request.
— hf-review
jrusso1020
left a comment
There was a problem hiding this comment.
Approval at exact head bdeb3672120b04c993eaede61fe35d9ce6d56977, unchanged since my code review. James directly authorized approval in the originating Slack thread; no new code findings were added in the refreshed review/comment history.
Package metadata, Context7 configuration and identical init agent templates read clean. Context7 schema validation passes and snapshot syntax matches the parser.
Validation and limits: Schema/source controls; no Context7 reindex, package publish, model evaluation or full init run.
Verdict: APPROVE
Reasoning: The reviewed code remains clear under the stated scope and collection contract; earlier blocking findings are resolved where applicable. No merge, deployment or publication performed by me.
— hf-review
# Conflicts: # packages/cli/package.json
jrusso1020
left a comment
There was a problem hiding this comment.
Re-approving at 5ca18fa after the merge from main. The PR diff against main has exactly the same added and removed lines as the diff I approved at bdeb367. The only differences are context and hunk offsets from the two resolved conflicts (version bump, template heading rename). Same 6 files and no new code. James authorized approval of this PR earlier, and that still applies.
What
hyperframes(the CLI package) gets the description "Make videos with code and AI agents: write HTML, render MP4. Open source." and 14 keywords. It had none.mcpis left out because the CLI ships no MCP server.npx hyperframes skills update, then ask for a video).context7.json: context7 (the docs index agents query) reads onlydocs,skills,registryandexamples. Today it also indexes 583 test-fixture files: the repo entry shows 21,755 snippets at "Medium" reputation. It also hands agents five composition rules taken from AGENTS.md. Claiming the library and merging our three context7 entries happens on context7.com, outside this PR.AGENTS.md/CLAUDE.mdthatinitwrites opens with one line saying what the folder is, and listssnapshot --at 1,3,5.Why
A clean Claude Code (no skills, no MCP, no user config) got 16 neutral "make a video" prompts over 20 runs. It picked HyperFrames 0 times and Remotion 3 times, and hand-built an HTML → headless Chrome → ffmpeg renderer 11 times. It ran no web searches, so it chose from memory and from what was on the machine. This PR won't change the memory part soon. It's cheap hygiene for the part we control: npm search and the files an agent reads in a project.
Proposed separately, needs repo admin:
gh repo edit heygen-com/hyperframes --add-topic claude-code --add-topic ai-agents --add-topic video-editor --add-topic codex --add-topic cursor --add-topic html-to-video --add-topic motion-graphics --add-topic remotion-alternativeNo visible change
Package metadata, one README line, the scaffolded agent notes and a context7 config. No UI.
Test plan
init.test.ts37/37