Add writing-style skill; keep the personal style file out of the mirror¶
- Date: 2026-10-21
- Status: Accepted
- Tags:
tooling|process
Context¶
A writing-style skill and /my-style command now live in ~/.claude, cloned from an open
source repo. The skill rewrites text in my voice and reads references/my-style.md, a profile
built from real Slack messages. A new rule in ~/.claude/CLAUDE.md makes any message I send
as myself load the skill first and default to a Slack draft.
syncing-harness-inventory mirrors ~/.claude into claude/harness/. The first snapshot
copied my-style.md into the mirror. It holds verbatim message text, channel names and
teammate names. Redaction covers employer identifiers, not personal prose, so it left them in.
Decision¶
- Record the
writing-styleskill and/my-stylecommand in the harness mirror. - Add
skills/writing-style/references/*toscripts/exclude.txt, so the style profile is dropped before redaction and never mirrored.
Alternatives Considered¶
| Option | Why Rejected |
|---|---|
| Mirror the style file and add redaction rules | Redaction cannot anonymize voice samples. The value of the file is the verbatim text. |
| Exclude the whole skill | Drops the reusable skill structure and the CLAUDE.md rule's target. |
| Leave it and rely on the repo being private | Visibility can change, and ADR-0038 already set the policy that personal prose is excluded. |
Consequences¶
Positive:
- The skill and command are tracked; my message text stays out of the repo.
Negative / Trade-offs:
- The mirror cannot reproduce my voice profile. It lives only in the skill's source repo checkout on disk.
Ongoing obligations:
- Any future personal reference file under a skill needs its own
exclude.txtline.
ADRs are immutable. If this decision changes, create a new ADR with status "Superseded by" pointing here, and update this record's status.