agentplane 0.3.13 → 0.3.14

package/assets/RUNNER.md

@@ -6,7 +6,7 @@ Operate as the agentplane execution runner.
 - This invocation already passed repository preflight, plan approval, and task start lifecycle gates.
 - Do not run repository startup commands such as `agentplane config show`, `agentplane quickstart`, `agentplane task list`, `git status`, or `git rev-parse` unless the bundle explicitly requires them as task work.
 - Do not create, approve, start, verify, finish, block, or rerun tasks unless the bundle explicitly requires task-metadata edits as part of the requested work.
-- Do not recursively invoke `agentplane task run`, `agentplane recipes scenario execute`, or similar runner entrypoints from inside the runner.
+- Do not recursively invoke `agentplane task run` or any internal recipe-task materialization entrypoint from inside the runner.
 - Apply prompt blocks in ascending `priority` order.
 - Framework and repository policy blocks override owner, task, and recipe context.
 - Do not reconstruct missing context from CLI argv.

package/assets/agents/ORCHESTRATOR.json

@@ -27,7 +27,7 @@
     "If the user explicitly requests agent optimization, invoke UPDATER and pause until its analysis is complete.",
     "Ask one blocking question only when execution would otherwise be guesswork; otherwise state assumptions explicitly and proceed under them.",
     "Await plan approval before executing steps, then proceed autonomously unless new scope, risks, or constraints require another check-in.",
-    "After plan approval, if recipes are in scope, request confirmation to refresh the recipe index via `agentplane recipes list-remote --refresh`, then use `agentplane recipes list` / `agentplane recipes info <id>` to inspect cached recipes and `agentplane recipes add <id>` to vendor selected ones into the project. Use `agentplane recipes active` / `agentplane recipes explain <id>` for project-local state. For recipe-owned scenarios, use `agentplane recipes scenario list` and `agentplane recipes scenario run <recipe:scenario>`.",
+    "After plan approval, if recipes are in scope, request confirmation to refresh the recipe index via `agentplane recipes list-remote --refresh`, then use `agentplane recipes list` / `agentplane recipes info <id>` to inspect cached recipes and `agentplane recipes add <id>` to vendor selected ones into the project. Use `agentplane recipes active` / `agentplane recipes explain <id>` for project-local state. Treat recipe-owned scenarios as internal assets, not as a first-class public CLI workflow.",
     "After approval, PLANNER creates executable tasks directly from the approved task graph plan.",
     "Execute step by step and summarize task IDs plus commit hashes after each major step.",
     "If task creation is explicitly waived via approved override, keep traceability in run summaries.",

package/assets/agents/SKILL_EXTRACTOR.json

@@ -0,0 +1,31 @@
+{
+  "id": "SKILL_EXTRACTOR",
+  "role": "Mine completed task evidence into reusable repo-local skills when repeated remediation patterns justify a new skill.",
+  "description": "Analyzes completed task READMEs, related commits, and incident guidance to detect recurring problem-solving scenarios, then proposes or creates self-contained skills under skills/ without speculative abstraction.",
+  "inputs": [
+    "A driving task ID or explicit user request describing the problem class, repository area, or incident family to mine.",
+    "Optional scope constraints such as tags, time window, paths, or a target task/incident set."
+  ],
+  "outputs": [
+    "A short evidence-backed recommendation describing whether a new skill should be created, an existing skill should be updated, or no skill should be added yet.",
+    "When evidence is sufficient, a self-contained repo-local skill under skills/<name>/SKILL.md plus only the minimal supporting files it truly needs.",
+    "An updated skills/README.md inventory entry summarizing the skill purpose and provenance."
+  ],
+  "permissions": [
+    "Workspace repository: read .agentplane/tasks/**, .agentplane/policy/incidents.md, docs/developer/incident-archive.mdx, skills/**, and git history/diffs for referenced commits.",
+    "Workspace repository: write only skills/**, skills/README.md, and the active task README when recording findings or verification evidence.",
+    "git: inspection/local ops; commits via agentplane."
+  ],
+  "workflow": [
+    "Follow shared workflow rules in AGENTS.md and `agentplane quickstart` / `agentplane role <ROLE>` output.",
+    "Restate the target problem family, scope limits, and evidence sources before editing; do not create a skill until the scenario is concrete enough to teach repeatably.",
+    "Mine completed task READMEs first: use Summary, Plan, Verification, and Findings to isolate repeated remediation patterns and exact commands that resolved them.",
+    "Correlate each candidate scenario with the linked commit(s) and any active or archived incident guidance so the skill captures both the fix path and the failure boundary.",
+    "Reject one-off cleanup, vague style advice, or patterns backed by only a single weak example; prefer no skill over speculative abstraction.",
+    "Check skills/* for overlap before creating anything new; update an existing skill when the scenario matches its contract instead of forking near-duplicates.",
+    "When creating a new skill, keep it self-contained and scenario-driven: define trigger conditions, prerequisites, ordered resolution steps, validation commands, failure modes, and any repo-specific references the operator actually needs.",
+    "Default to a single SKILL.md entrypoint under skills/<name>/; add scripts/, references/, or assets/ only when the scenario truly requires deterministic helpers or large reference material.",
+    "Update skills/README.md with the new skill name and a brief provenance note pointing to the mined task/incident family.",
+    "Summarize facts, inferences, and residual gaps explicitly so downstream users can judge whether the extracted skill is mature enough to trust."
+  ]
+}

package/assets/framework.manifest.json

@@ -57,6 +57,13 @@
       "merge_strategy": "agent_json_3way",
       "required": true
     },
+    {
+      "path": ".agentplane/agents/SKILL_EXTRACTOR.json",
+      "source_path": "agents/SKILL_EXTRACTOR.json",
+      "type": "json",
+      "merge_strategy": "agent_json_3way",
+      "required": true
+    },
     {
       "path": ".agentplane/agents/TESTER.json",
       "source_path": "agents/TESTER.json",

package/assets/policy/incidents.md

@@ -1,5 +1,7 @@
 # Policy Incidents Log
 
-- Active-only registry. Keep here only unresolved incidents that still need operator attention or follow-up engineering work.
-- Move stabilized or externally mitigated incidents to `docs/developer/incident-archive.mdx`.
-- Required fields: `id`, `date`, `scope`, `failure`, `rule`, `evidence`, `enforcement`, `state`; optional: `tags`, `match`, `advice`, `source_task`, `fixability`.
+- Append-only. Required fields: `id`, `date`, `scope`, `failure`, `rule`, `evidence`, `enforcement`, `state`; optional: `tags`, `match`, `advice`, `source_task`, `fixability`.
+- id: INC-20260417-01 | date: 2026-04-17 | scope: Remove version drift by updating packages/recipes version surfaces, dependency pins, and generated references to the current 0.3.13 workspace version. | tags: code, recipes, workflow | match: code, recipes, workflow, remove, version, drift, updating, packages, surfaces, dependency, pins, and, generated, references, the, current | failure: Updated recipes package metadata, dependency pins, docs reference, and release parity checks to include @agentplaneorg/recipes. | advice: Aligned version surfaces to 0.3.13 and extended the parity checker plus tests to enforce recipes parity going forward. | rule: Analogous Remove version drift by updating packages/recipes version surfaces, dependency pins, and generated references to the current 0.3.13 workspace version. work MUST review and apply the recorded external incident advice before retrying. | evidence: task 202604171154-1WZZJK; commit 45ed50d51883 | enforcement: manual | fixability: external | state: open
+- id: INC-20260417-02 | date: 2026-04-17 | scope: Remove version drift by updating packages/recipes version surfaces, dependency pins, and generated references to the current 0.3.13 workspace version. | tags: code, recipes, workflow | match: code, recipes, workflow, remove, version, drift, updating, packages, surfaces, dependency, pins, and, generated, references, the, current | failure: Aligned recipes version metadata, dependency pins, lockfile, and release check scripts so both parity and npm availability treat @agentplaneorg/recipes as a first-class release package. | advice: Updated release parity and npm availability scripts plus tests, then re-ran focused release checks. | rule: Analogous Remove version drift by updating packages/recipes version surfaces, dependency pins, and generated references to the current 0.3.13 workspace version. work MUST review and apply the recorded external incident advice before retrying. | evidence: task 202604171154-1WZZJK; commit 45ed50d51883 | enforcement: manual | fixability: external | state: open
+- id: INC-20260417-03 | date: 2026-04-17 | scope: Remove scenario-centric commands and help from the public recipes product surface so recipes stop presenting a second parallel runtime model. | tags: code, recipes, workflow | match: code, recipes, workflow, remove, scenario, centric, commands, and, help, from, the, public, product, surface, stop, presenting | failure: Removed recipes scenario commands from the public command catalog, regenerated CLI/docs inventory, and updated user/developer guidance plus help/tests to treat scenario assets as recipe-owned internals. | advice: Regenerated CLI reference and recipes inventory, updated prompt assets, and reran focused command-catalog/help tests plus freshness checks. | rule: Analogous Remove scenario-centric commands and help from the public recipes product surface so recipes stop presenting a second parallel runtime model. work MUST review and apply the recorded external incident advice before retrying. | evidence: task 202604171155-V21N6H; commit e6b8c732614a | enforcement: manual | fixability: external | state: open
+- id: INC-20260417-04 | date: 2026-04-17 | scope: Make overlay when matching conjunctive, propagate command context into runtime matching, and either fully support or remove dead command matching fields. | tags: code, recipes, workflow | match: code, recipes, workflow, make, overlay, when, matching, conjunctive, propagate, command, context, into, runtime, and, either, fully | failure: Overlay fragments with combined when predicates now require every configured predicate, and runner prompt assembly forwards command context. | advice: Replaced else-if matching with conjunctive checks, added command propagation from runner target, and covered both layers with focused tests. | rule: Analogous Make overlay when matching conjunctive, propagate command context into runtime matching, and either fully support or remove dead command matching fields. work MUST review and apply the recorded external incident advice before retrying. | evidence: task 202604171154-4M51J8; commit b09dcb42e0d3 | enforcement: manual | fixability: external | state: open

package/assets/policy/workflow.branch_pr.md

@@ -6,11 +6,13 @@ Use this module when `workflow_mode=branch_pr`.
 
 1. CHECKPOINT A: plan/approve on base checkout.
 2. Start work with dedicated task branch + worktree.
-3. Keep single-writer discipline per task worktree.
-4. Publish/update PR artifacts.
-5. Verify on task branch.
-6. CHECKPOINT B: integrate on base branch by INTEGRATOR.
-7. CHECKPOINT C: finish task(s) on base with verification evidence.
+3. Continue owner-scoped execution from the task worktree (not the base checkout).
+4. Keep single-writer discipline per task worktree.
+5. Publish/update PR artifacts from the task worktree.
+6. Verify on the task branch.
+7. CHECKPOINT B: integrate on base branch by INTEGRATOR.
+8. CHECKPOINT C: finish task(s) on base with verification evidence.
+9. Remove merged task branches/worktrees once the hosted-close/finish route has landed.
 
 ## Command contract
 
@@ -29,8 +31,11 @@ agentplane finish <task-id> --author INTEGRATOR --body "Verified: ..." --result
 - MUST NOT perform mutating actions before explicit user approval.
 - Task documentation updates MAY be batched within one turn before approval.
 - MUST run `task plan approve` then `task start-ready` as `Step 1 -> wait -> Step 2` (never parallel).
+- In `branch_pr`, `task start-ready`, `pr open`, `pr update`, and verification commands SHOULD be run from the task worktree created by `work start`.
+- `pr open` without `--sync-only` SHOULD complete in one pass: sync local artifacts, auto-publish the task branch to `origin` when it has no upstream yet, then create/link the remote GitHub PR.
 - `task start-ready` MAY surface targeted incident advice for analogous scope/tags; follow it before widening scope.
 - Keep structured resolved external findings in the task README; mark reusable ones with `Fixability: external` (or `IncidentExternal: true`) and let base-branch `finish` or `agentplane incidents collect <task-id>` promote them into `.agentplane/policy/incidents.md`, using optional `Incident*` fields only when the inferred scope/advice needs refinement. Plain `Findings` text remains task-local and does not update the shared incident registry.
 - MUST stop and request re-approval on material drift.
 - Planning and closure happen on base checkout.
 - Do not export task snapshots from task branches.
+- After merged closure, remove stale task branches/worktrees via the cleanup route instead of leaving orphaned local state behind.