Hyphenomenon Chat Intake

Hyphenomenon Chat Intake

Custom skill created by Maggie Lerman.

Create a durable, transcript-backed Hyphenomenon chat intake artifact.

This skill covers chat/session/conversation intake. For repository/project dossiers, use hyphenomenon-project-intake.

Workflow

1. Orient in the Hyphenomenon repo.

• Read AGENTS.md, CHANGELOG.md, ROADMAP.md, DOCS/intake/README.md, DOCS/features/chats.md, and scripts/intake/chat/README.md.
• Keep project 1059 in mind for chat intake contract changes unless a newer project supersedes it.

2. Capture the raw transcript.

• Store every new intaked chat transcript in git under DOCS/intake/runs/chats/<date>-<slug>/transcript.txt.
• Historical transcripts under scripts/intake/chat/runs/transcripts/ remain valid legacy records, but do not use that path for new runs.
• The private transcript archive must be a complete source record, not a hand-written summary or curated retelling.
• Preserve the full turn/event sequence. Include tool calls and tool outputs in parseable form; summarize tool-output sections only when raw output is too large/noisy, and mark the section with raw-output size/hash or equivalent provenance.
• Mark every redaction explicitly. If there are no redactions, say so in the archive.
• For Codex local sessions, prefer the repo helper npm run intake:chat:transcript:codex-jsonl -- --jsonl "/abs/path/to/rollout-...jsonl" --out "/abs/path/to/transcript.txt" --thread-id "<id>", which writes source metadata plus the complete JSONL event stream.
• For non-Codex exports, preserve speaker turns in parseable text such as User:, Assistant:, Tool:, and System:.
• Do not commit inline raw transcript text in JSON. Use transcriptSourcePath.

3. Build the JSON artifact.

• Save the run artifact under DOCS/intake/runs/chats/<date>-<slug>/artifact.json.
• Set transcriptSourcePath relative to the artifact file, normally transcript.txt.
• Set transcriptReviewedAt after reviewing/redacting the transcript.
• Keep public content curated: summary, highlights, decisions, contentBodyMarkdown, verified sourceExtracts, optional conversationRender.featuredMoment, and public-safe links.
• Public summaries belong only in the JSON artifact/public fields; never use a public summary as the private transcript archive.
• sourceExtracts[].excerpt and conversationRender.turns[].content must be verbatim public-safe transcript excerpts, not summaries.

4. Capture created work.

• Use createdArtifacts for reusable artifacts created in the chat.
• Supported type values are custom_skill, project, and repository.
• Use relatedProjects for existing or newly created Hyphenomenon project context, with created_from_conversation or created_from_decision when appropriate.
• Keep private repo paths out of public link fields; href must be a site path or public non-GitHub URL.

5. Use screenshots only when they are real evidence.

• Real screenshots are good when they show an artifact, UI state, rendered route, diagram, product surface, or external visual proof created or discussed in the chat.
• Do not generate images of transcript text, source text, or quote cards as a substitute for provenance.
• If no real visual artifact exists, omit featureImage and evidenceImages.

6. Validate and sync.

npm run intake:validate:chat -- --artifact "DOCS/intake/runs/chats/<date>-<slug>/artifact.json"
npm run intake:sync:chat -- --artifact "DOCS/intake/runs/chats/<date>-<slug>/artifact.json"
npm run intake:sync:chat:apply -- --artifact "DOCS/intake/runs/chats/<date>-<slug>/artifact.json"

7. Verify the rendered route.

• Check /chats/<slug> after apply.
• Confirm the page renders curated public-safe content, optional real screenshots, related projects, and created artifacts.
• Confirm the page does not render or link the raw transcript by default.

8. Commit and push intake records.

• Commit the JSON artifact and transcript archive together.
• Commit optional real screenshot media only when it is public-safe and relevant.
• Push the branch when the user wants remote-backed transcript retention.