← ClaudeAtlas

p2a-nextlisted

Use when a Plan2Agent user asks what to do next, wants to resume a project, or needs one state-based next action in an agent session.
silbaram/plan2agent · ★ 0 · AI & Automation · score 72
Install: claude install-skill silbaram/plan2agent
# Plan2Agent Next Use the CLI result as the only decision authority. Do not infer, list, or encode project-state rules in this skill. ## Procedure 1. Run `p2a next --json --contract v2` from the target project. Pass `--project-id <id>` only when the user selected an artifact explicitly. The unqualified JSON command `p2a next --json` remains the backward-compatible `p2a.next.v1` surface; human output defaults to v2 so structured options remain visible. 2. Parse the one returned `command`, `continuation`, `reason`, `reasonCode`, and `state`. `reasonCode` is the stable schema-enumerated machine identifier and currently equals `state` for compatibility. 3. Present the result as a short development-assistant message: what P2A understood, the one material decision needed (if any), what it will do next, and why that action is recommended. Do not expose Gate names, state ids, run ids, hashes, or artifact paths unless the user asks for details or troubleshooting. Keep the exact returned command and reason available internally; never alter their machine meaning. If the v2 result contains an `immediate` continuation, request its references once with `p2a context show --artifacts <dir> --continuation <id> --provider <provider>`. For `after_command_success`, first validate the single `p2a.execution_result.v1` result, then pass its `runId` to the same command. Do not rediscover or individually reopen packet-managed references. ## Handle the returned command - For `kind: cli` with `re