Native commands inside Coding/Claw
| Command | Purpose |
|---|---|
/goal or /goal status |
Human-readable current Goal, next actor and distinct runtime facts. |
/goal create ... |
Create the owner's current Goal; no worker is started. |
/goal pause ... |
Stop proactive continuation/owned observation. |
/goal resume |
Explicitly resume a paused Goal under actual authority. |
/goal clear ... |
Clear the binding without undoing accepted effects or deleting task history. |
/goal history |
Read historical Goals. |
/goal status --json |
Explicit full machine-readable status. |
Commands use the same owner-bound service as model tools. No extra Goal enable switch or automatic Goal creation is needed.
Model workflow
| Operation | Owner of the decision |
|---|---|
| create / update / plan | Current main Agent, using its real conversation and user intent. |
| report | Main Agent assesses baseline/advanced/unchanged/regressed with evidence and next steps. |
| tasks / task_update | Authoritative task data; children need explicit assigned-task scope. |
| wait | Main Agent waits for tasks or a specific user question; not for itself. |
| activate | Return from an eligible wait to active; it cannot wash away user authority requirements. |
| pause / resume | Explicit pause/resume semantics; no timer-based permission release. |
| complete | Main Agent's evidence-based judgment, not a script/reviewer verdict. |
| status / history | Reads, not proof of model execution or channel delivery. |
active/waiting/paused/completed Goal state is separate from Agent execution
and reminder delivery. Task waits can react to relevant changes; a completed Goal
remains visible until explicit clear or a new Goal. Use the actual tool schema for action-specific fields;
omit unrelated parameters rather than filling null placeholders.
Files and boundaries
Tasks live in tasks/*.json, plans in plan.md, progress in progress.md and the
optional generated observation in eyes.json under the Goal workspace. Model
workers should use their assigned scoped Goal tool. A low-level task_cli CAS
helper exists for explicit external integrations, not as a required child path.
plan is also a packaged instruction skill, not a second planning engine. Neither
skill activation nor Goal reminders grant tool permission. See the
package guide
for model action schemas, persistence and restart limits.