1. Treat the first word as $ACTION, remainder as $REST.
2. With no input, or input that is neither an action nor a follow-up on a change, show help and stop.
3. For a valid action, read its reference to interpret $REST.
4. Run the init gate, except for explore.
5. For continue, apply, verify, archive and follow-ups, select the change <name>.
6. Follow the action's instructions, or for a conversational follow-up, run node <skill-dir>/scripts/cli.mjs status "<name>" --apply and read the affected artifacts and code.

Init Gate

Run node <skill-dir>/scripts/cli.mjs init --status:

• If uninitialized and $ACTION is not init, suggest /esdd init and stop.
• If initialized and $ACTION is init, report that and stop.

Select the Change

• Use the named change, then the one established in conversation. Otherwise, run node <skill-dir>/scripts/cli.mjs list and use the only relevant open change; ask if several are plausible.
• If none exist, report and stop; suggest /esdd new for planning or implementation requests.
• If added scope could extend the current change or start another, ask which unless already clear.

Working Rules

Actions may further restrict edits.

• Discussion and review do not authorize edits; completion does not authorize the next phase.
• Use revision for artifact corrections before changing code that depends on them.
• Correct authorized code against the agreed target. Remove abandoned code within scope and preserve unrelated work. Keep repair details in the conversation or subagent handoff, not artifacts. Code corrections need neither unchecked groups nor /esdd apply.
• Check tasks only when implementation and relevant verification satisfy their outcomes; uncheck unmet ones. Run relevant tests and build/type checks after code corrections.

Action References

• init
• document
• explore
• new
• continue
• apply
• verify
• archive
• view
• help