| name | voice-enrich |
| description | Use when enriching a person profile with voice transcript data and Perel-Baldwin relational writing should be integrated into the note update.
|
| argument-hint | <name> [--mode rewrite|review|create] [--analyst-report /abs/path.md] |
| allowed-tools | Read, Write, Edit, Glob, Grep, Bash(bun run *), Agent |
| disable-model-invocation | true |
Voice Enrich
Use this when Nathan wants Perel-Baldwin involved in a people-note workflow.
This skill keeps /people-enrich lean:
- research stays upstream
- voice stays with
perel-baldwin
- note mutation stays with
apply-enrichment.ts
Defaults
- If
--mode is omitted and a unique existing enrichment report can be resolved in runtime/people-enrichment/tmp/, use rewrite
- If
--mode is omitted and no existing enrichment report exists, use review
create must be explicit
Required Inputs
Always load:
~/code/my-second-brain/context/people/nathan-vale.md- target person note at
~/code/my-second-brain/context/people/<slug>.md
@context/contract-perel-baldwin-context.md
Optional:
~/code/my-second-brain/context/context/personal.md- an explicit
--analyst-report /absolute/path.md
If the target person note does not exist, stop. This skill is for people-note workflows, not note creation from nothing.
Mode Selection
Rewrite
Use when an existing enrichment report already exists at:
- a resolved path inside
~/code/my-second-brain/runtime/people-enrichment/tmp/
Resolution order:
- exact
~/code/my-second-brain/runtime/people-enrichment/tmp/<slug>-enrichment-report.json
- otherwise, a unique
*-enrichment-report.json candidate that clearly matches the target name
If multiple candidates match, stop and ask Nathan which artifact to use.
Load output contract:
@context/contract-people-note.md
Review
Use when Nathan wants critique and suggested rewrites without running the writer.
Load output contract:
@context/contract-people-note-review.md
Create
Use only when Nathan explicitly asks for a fresh EnrichmentReport and provides upstream evidence beyond the note itself.
Good create-mode evidence:
--analyst-report /absolute/path.md- explicit QMD findings
- explicit psychometrics
- explicit thread or argument summaries
If there is no substantive evidence beyond the note, stop and explain that create mode needs upstream evidence before it is worth running.
Load output contract:
@context/contract-people-note-create.md
Workflow
Step 1 - Run The Helper
The operational entrypoint is the thin Bun helper:
bun run ~/.claude/skills/voice-enrich/scripts/voice-enrich.ts <name> [--mode rewrite|review|create] [--analyst-report /absolute/path.md]
What the helper owns:
- resolve the target note deterministically
- select or validate the mode
- resolve the existing enrichment report for rewrite
- build the ContextBundle in deterministic order
- save the bundle artifact for inspection
- dispatch
perel-baldwin with the bundle as the full allowed context
- save preview artifacts in
runtime/people-enrichment/tmp/
- run
apply-enrichment.ts only for JSON-producing modes
Step 2 - Read The Produced Artifacts
The helper prints machine-readable paths. Read the relevant outputs:
- rewrite:
<stem>-voiced-report.json and <stem>-voiced-proposed.md
- review:
<slug>-review.md
- create:
<slug>-created-report.json and <slug>-created-proposed.md
Step 3 - Present The Preview
- For rewrite/create, present a concise diff or change summary from the proposed markdown
- For review, present the markdown review directly
- Do not write the live note without explicit approval
Safety Rules
- Never bypass the shared writer for JSON-producing modes
- Never write the live note without explicit approval
- Never treat
analyst-report as stronger than direct note/report evidence
- Never invent create-mode depth from the note alone
- Keep preview artifacts in
runtime/people-enrichment/tmp/
