By name: write
description: "Top-level write orchestrator: outline a new page, get user approval, then draft from the outline."
argument-hint: "[--non-interactive] [--no-log] "
user-invocable: true
/inki:write: outline then draft
Step 0: Parse arguments
If $ARGUMENTS contains --help or -h, print usage and stop, per ../../references/help.md. Do not run the workflow.
Otherwise, from $ARGUMENTS, detect the autonomy flag anywhere in the list: --non-interactive (canonical), aliases --auto-approve, --auto, --yes, -y, --no-questions-asked (all equivalent). If present, set AUTO=true and remove the flag. What remains is the topic brief (text or path to a .md file).
Logging: unless --no-log is passed, write this skill's report to the run log per ../../references/logging.md (--log-dir <path> and --short-log are also accepted). When invoked as part of an orchestrator (e.g. /inki:document), write into that run's existing directory instead of creating a new one.
Workflow
Interactive (AUTO=false, default)
- Invoke
/inki:outline $ARGUMENTS. Wait for user approval inside that sub-skill. - Once an outline file exists and the user accepted it, invoke
/inki:draft <outline-path>.
If the user rejects the outline, stop. Do not draft.
Auto (AUTO=true)
- Invoke
/inki:outline --non-interactive $ARGUMENTS. The outline is generated and saved without an approval gate. - Immediately invoke
/inki:draft <outline-path>on the resulting outline.
In auto mode, the user sees the outline and the draft only after both are produced. They can still discard the outputs if they don't fit; nothing is committed automatically. --non-interactive here means "don't pause between outline and draft," not "trust the output blindly."
Rules
- Even in auto mode, do not skip the outline step. The outline must exist as a file before drafting.
- Do not invent facts beyond what the brief and template support.
- Do not commit, push, or open a PR. Use
/inki:submitfor that.