codexkit-follow-up-draft-builder
Draft concise follow-up messages, nudges, reminders, and recap emails from open decisions, stale requests, or recent meetings. Use when the work is repetitive business writing with known context. Do n
SKILL.mdskills/codexkit-follow-up-draft-builder/SKILL.md
GitHubFollow-Up Draft Builder
Purpose
Write short operational follow-ups that move work forward without overthinking.
When to use
- someone needs a reminder, nudge, or recap
- open decisions need a clean ask and due date
- the task is routine communication, not persuasion strategy
When not to use
- the message carries legal, HR, or regulatory risk
- tone calibration matters more than speed
Inputs
- source context or prior thread summary
- target audience
- goal of the message
- preferred tone if known
Procedure
- Identify the ask, owner, and desired timing.
- Strip background detail down to only what the receiver needs.
- Draft the message with a clear subject or opening line.
- End with one explicit next step.
- Provide a shorter variant if the user likely needs chat or Slack format.
Output
- follow-up message draft
- optional short-form version
Definition of done
- the request is clear in one read
- the recipient knows what action is needed
- the draft is brief enough to send with minor edits
Examples
- "Write a reminder to finance for the missing close evidence."
- "Draft a polite follow-up to the vendor on the contract intake questions."
Quality Criteria
- Trigger conditions and input requirements are unambiguous
- Each automated step produces a verifiable output
- Error handling and fallback paths are defined
- Manual override points are documented
Verification (4C)
| Check | Question |
|---|---|
| Correctness | Does the workflow produce the expected output for the defined inputs? |
| Completeness | Are all trigger conditions, edge cases, and error paths handled? |
| Context-fit | Is the automation appropriate for the frequency and criticality of this task? |
| Consequence | If this ran unattended and failed silently, what would the downstream impact be? |
Edge Cases
- Input format varies unexpectedly — Add a normalization step at entry. Alert the operator on format mismatches.
- Downstream system is unavailable — Queue the output and retry with exponential backoff. Alert after N failures.
- Partial execution completes — Ensure idempotency — re-running from the start produces the same result without duplication.
Changelog
- v1.0.0 — Initial release