Write Like Me Bootstrap
Create or update a repo-local writing-style skill from the user's real authored
messages. The output should help future assistants draft in the user's voice
without copying raw private messages into durable files.
Default Output
Write the generated skill to .codex/skills/write-like-me/ unless the user asks
for a different name. Use:
.codex/skills/write-like-me/SKILL.md for the durable drafting workflow.codex/skills/write-like-me/references/style-profile.md for compact voice, posture, and examples.codex/skills/write-like-me/agents/openai.yaml if the repo uses skill UI metadata
Use references/style-profile-template.md as the target shape.
Use references/generated-skill-template.md as the starting shape for the
generated SKILL.md.
Source Scan
Use connected Slack and email sources when available. Prefer authored messages:
- Slack messages sent by the user in channels, DMs, and group DMs
- sent email, replies, follow-ups, scheduling notes, intros, escalations, and updates
- recent enough writing to reflect the current voice, usually the last 90-180 days
- older writing only when it clarifies stable style or an important posture
If a connector is unavailable, say what is missing and proceed with available
sources or ask whether to retry later.
Privacy Boundary
Do not save raw Slack or email excerpts in durable files. Use short synthetic
examples that preserve style without exposing private content, names, deals,
secrets, or account details. Keep source evidence as compact descriptors such as
sent email follow-ups, March-June 2026 or Slack replies in project channels.
Ask before writing or updating the generated skill. Reading connected context for
this bootstrap is allowed only when the user has asked for onboarding or this
skill, and writing still requires explicit approval.
Analysis Workflow
- Identify the user's authored messages and separate them by channel and intent.
- Cluster writing into postures rather than one generic voice. Useful default
postures:
- quick Slack reply
- decision or pushback
- delegation or ask
- executive/status update
- email reply
- intro or relationship note
- scheduling/admin note
- apology, correction, or repair
- For each posture, infer:
- when to use it
- sentence length and pacing
- directness, warmth, humor, and formality
- how the user frames asks, tradeoffs, disagreement, and urgency
- phrases or moves to reuse as patterns, not verbatim quotes
- things the user avoids
- Compare Slack and email. Preserve differences instead of flattening them.
- Draft the generated
write-like-me skill and profile.
- Present a concise preview of the inferred postures and ask for approval before
writing files.
- After approval, create or update the generated skill in this repo.
Generated Skill Requirements
The generated write-like-me skill should:
- trigger for drafting, rewriting, critiquing, or replying in the user's voice
- route by channel and posture before drafting
- ask only when audience, channel, or goal would materially change the draft
- default to concise, useful drafts rather than style analysis
- include a critique mode that says what feels unlike the user
- preserve privacy by never revealing the source messages used for bootstrapping
- tell future assistants to update the profile when the user corrects voice,
posture, or audience assumptions
Onboarding Behavior
During Assistant onboarding, suggest this bootstrap after the first connected
Slack/email scan finds enough authored writing to infer voice. The offer should
be concrete:
I can also bootstrap a write-like-me skill from your sent email and Slack messages, split by posture like quick replies, pushback, delegation, intros, and status updates. Want me to do that?
If the user says yes, run the source scan, show the posture preview, and ask
before writing the generated skill.
1---2name: write-like-me-bootstrap3description: Bootstrap a durable "write like me" skill from the user's own Slack and email writing. Use during Assistant onboarding, after Slack/Gmail or email connectors are available, or when the user asks to learn their voice, infer writing style, create a personal writing persona, generate a write-like-me skill, analyze sent messages, or capture different writing postures across email and Slack.4---56# Write Like Me Bootstrap78Create or update a repo-local writing-style skill from the user's real authored9messages. The output should help future assistants draft in the user's voice10without copying raw private messages into durable files.1112## Default Output1314Write the generated skill to `.codex/skills/write-like-me/` unless the user asks15for a different name. Use:1617- `.codex/skills/write-like-me/SKILL.md` for the durable drafting workflow18- `.codex/skills/write-like-me/references/style-profile.md` for compact voice, posture, and examples19- `.codex/skills/write-like-me/agents/openai.yaml` if the repo uses skill UI metadata2021Use `references/style-profile-template.md` as the target shape.22Use `references/generated-skill-template.md` as the starting shape for the23generated `SKILL.md`.2425## Source Scan2627Use connected Slack and email sources when available. Prefer authored messages:2829- Slack messages sent by the user in channels, DMs, and group DMs30- sent email, replies, follow-ups, scheduling notes, intros, escalations, and updates31- recent enough writing to reflect the current voice, usually the last 90-180 days32- older writing only when it clarifies stable style or an important posture3334If a connector is unavailable, say what is missing and proceed with available35sources or ask whether to retry later.3637## Privacy Boundary3839Do not save raw Slack or email excerpts in durable files. Use short synthetic40examples that preserve style without exposing private content, names, deals,41secrets, or account details. Keep source evidence as compact descriptors such as42`sent email follow-ups, March-June 2026` or `Slack replies in project channels`.4344Ask before writing or updating the generated skill. Reading connected context for45this bootstrap is allowed only when the user has asked for onboarding or this46skill, and writing still requires explicit approval.4748## Analysis Workflow49501. Identify the user's authored messages and separate them by channel and intent.512. Cluster writing into postures rather than one generic voice. Useful default52 postures:53 - quick Slack reply54 - decision or pushback55 - delegation or ask56 - executive/status update57 - email reply58 - intro or relationship note59 - scheduling/admin note60 - apology, correction, or repair613. For each posture, infer:62 - when to use it63 - sentence length and pacing64 - directness, warmth, humor, and formality65 - how the user frames asks, tradeoffs, disagreement, and urgency66 - phrases or moves to reuse as patterns, not verbatim quotes67 - things the user avoids684. Compare Slack and email. Preserve differences instead of flattening them.695. Draft the generated `write-like-me` skill and profile.706. Present a concise preview of the inferred postures and ask for approval before71 writing files.727. After approval, create or update the generated skill in this repo.7374## Generated Skill Requirements7576The generated `write-like-me` skill should:7778- trigger for drafting, rewriting, critiquing, or replying in the user's voice79- route by channel and posture before drafting80- ask only when audience, channel, or goal would materially change the draft81- default to concise, useful drafts rather than style analysis82- include a critique mode that says what feels unlike the user83- preserve privacy by never revealing the source messages used for bootstrapping84- tell future assistants to update the profile when the user corrects voice,85 posture, or audience assumptions8687## Onboarding Behavior8889During Assistant onboarding, suggest this bootstrap after the first connected90Slack/email scan finds enough authored writing to infer voice. The offer should91be concrete:9293```text94I can also bootstrap a write-like-me skill from your sent email and Slack messages, split by posture like quick replies, pushback, delegation, intros, and status updates. Want me to do that?95```9697If the user says yes, run the source scan, show the posture preview, and ask98before writing the generated skill.
jxnl