@unbrained/pm-cli 2026.9.6 → 2026.9.8

package/.agents/skills/pm-developer/SKILL.md

@@ -41,8 +41,9 @@ pm guide workflows --depth brief
 ## Non-Negotiables
 
 - Author identity is detected automatically. **Never pass `--author` and never
-  set `PM_AUTHOR`.** The harness, model, effort, role, and topic are probed and
-  recorded on every history entry.
+  set `PM_AUTHOR`.** Supported harness signals can supply model, effort, role,
+  and topic. Missing observations remain unknown; do not infer them from a
+  configured default or invent provenance for historical entries.
 - Never edit files under `.agents/pm` directly. Every mutation goes through `pm`.
 - Never claim an item's state from memory. Read it live before asserting it.
 - Claim before substantial edits; release when paused, handed off, or closed.
@@ -62,10 +63,17 @@ pm test  <ID> --add command="node scripts/run-tests.mjs test -- <target>",scope=
 pm comments <ID> "Evidence: <what changed and what passed>"
 pm learnings <ID> --add "<lesson that outlives this item>"
 pm test <ID> --run --progress
-pm close <ID> "<reason with evidence>" --validate-close warn
+pm close <ID> "<reason with evidence>" --resolution "<delivered change>" --expected "<required outcome>" --actual "<observed outcome>" --validate-close warn
 pm release <ID>
 ```
 
+Before closing a repository defect, inspect its declared `escape_class` and
+`gate_evidence` contract and record the required structured evidence. In pm-cli,
+follow [Defect Evidence](../../../docs/DEFECT_RECURRENCE.md#defect-evidence-on-pm-items)
+and run `pnpm quality:defect-evidence` **after** closure: a gate run while the item
+is active cannot prove terminal-only requirements. Verify runner changes with
+both build-enabled and prebuilt CI environments.
+
 ## Capability Map
 
 Capability families and their commands are generated from the public SDK
@@ -78,7 +86,7 @@ enters, and use these drift-gated references:
 `pm help` shows the bounded core tier. Use `pm guide capabilities`, `pm
 contracts --summary`, or `pm <command> --help --json` for progressive
 disclosure into the wider surface. Durable lessons and history maintenance
-(`pm learnings`, `pm history --verify`, `history-repair`, `history-redact`,
+(`pm learnings`, `pm history <ID> --verify`, `history-repair`, `history-redact`,
 `history-compact`, `close-many`) are routed by `pm guide evidence`.
 
 ## Token Discipline

package/.agents/skills/pm-developer/references/MULTI_AGENT_MERGE.md

@@ -19,15 +19,16 @@ Skipping this is the single most common cause of tracker merge conflicts.
 
 ## Ownership Across Agents
 
-- `pm claim <ID>` before substantial edits. A claim is a recorded lease, not a
-  lock file, and it survives branching.
+- `pm claim <ID>` before substantial edits. Claim provenance is recorded with
+  the item and must be reconciled across branches.
 - `pm release <ID>` when pausing, handing off, closing, or canceling.
-- `pm list --status in_progress` shows what the fleet currently holds.
+- `pm list --status in_progress` shows work marked as executing. A claim alone
+  does not set that status; inspect `pm get <ID>` for its current claim state.
 - Never force an ownership or lock override without explicit human approval.
 
-Author identity is detected per invocation — harness, model, effort, role, and
-topic — so two agents on the same branch remain distinguishable in history
-without anyone passing `--author`.
+Author identity is detected per invocation from available harness signals.
+Model, effort, role, and topic observations are retained when available; their
+absence is not evidence of a particular model or role. No `--author` is needed.
 
 ## What Merges Field-Aware
 
@@ -51,8 +52,12 @@ reconciliation. Always run the post-merge gate:
 
 ```bash
 pm merge report
+pm merge reconcile --dry-run --json
+# After reviewing the receipt, apply the reconciliation.
 pm merge reconcile --message "Post-merge history reconciliation"
-pm validate --check-history-drift
+pm validate --check-storage-integrity --check-history-drift
+pm history <ID> --verify --strict-exit
+pm history _workspace --verify --strict-exit
 pm health --check-only
 ```
 
@@ -67,6 +72,10 @@ pm graph audit
 pm activity --limit 20
 ```
 
-History is append-only and hash-chained. `--check-history-drift` compares each
-stream's recorded chain against its contents, so a silently rewritten entry
-fails the check rather than passing unnoticed.
+Ordinary mutations append hash-chained history. `--check-history-drift` checks
+chain/replay agreement, and `history <ID> --verify --strict-exit` verifies a
+specific stream. These checks detect corruption and inconsistent rewrites;
+they do not independently prove that an attacker never replaced a whole chain
+and recomputed its hashes. Detached attestations or externally anchored
+receipts provide that separate proof boundary. Never rewrite historical
+evidence merely to make a graph or completeness metric look better.

package/.agents/skills/pm-extensions/SKILL.md

@@ -99,6 +99,12 @@ flag contract for each one.
 
 ## References
 
+Use the [generated capability routing](../../../docs/generated/AGENT_CAPABILITY_ROUTING.md)
+to discover the complete contract-owned surface. A capability that does not
+suit this project's real workflow should have a sandbox scenario or an explicit
+waiver on its canonical item; creating artificial production data is not proof
+of useful adoption.
+
 | Need                              | Load                                                | Cost     |
 | --------------------------------- | ---------------------------------------------------- | -------- |
 | Lifecycle recipes                  | [Extension lifecycle](references/LIFECYCLE.md)       | ~450 tok |

package/.agents/skills/pm-sdk/SKILL.md

@@ -57,8 +57,10 @@ proportional to what you import.
 Query the shipped surface instead of guessing a symbol name:
 
 ```bash
-jq -r '.entrypoints["./sdk/graph"].symbols[].name' \
-  node_modules/@unbrained/pm-cli/sdk/public-surface.json | head -40
+node --input-type=module -e '
+  import surface from "@unbrained/pm-cli/sdk/public-surface.json" with { type: "json" };
+  console.log(JSON.stringify(surface));
+' | jq -r '.entrypoints["./sdk/graph"].symbols[:40][].name'
 ```
 
 ## Non-Negotiables
@@ -99,6 +101,13 @@ so a new public entrypoint cannot ship ungoverned.
 
 ## References
 
+The [generated capability routing](../../../docs/generated/AGENT_CAPABILITY_ROUTING.md)
+covers every declared family. Expand only the family needed, then query its
+published SDK symbols and runtime contract. For MCP integrations, start with
+the [current conformance matrix](../../../docs/MCP_2026_07_28_CONFORMANCE.md)
+and its SDK-owned request, transport, task, subscription, Skills, and Apps
+contracts; distinguish implementation evidence from exact-release acceptance.
+
 | Need                                        | Load                                                    | Cost     |
 | ------------------------------------------- | -------------------------------------------------------- | -------- |
 | Step-by-step integration checklist           | [Integration checklist](references/INTEGRATION_CHECKLIST.md) | ~400 tok |

package/.agents/skills/pm-user/SKILL.md

@@ -78,6 +78,10 @@ the criteria; `--dep` **appends**.
 
 ## Capability Map
 
+The complete, drift-gated [capability routing](../../../docs/generated/AGENT_CAPABILITY_ROUTING.md)
+is generated from SDK contracts. This table routes common workflows; use
+`pm guide capabilities` for the whole installed surface.
+
 | Need                        | Entry                                    | Guide topic  |
 | --------------------------- | ---------------------------------------- | ------------ |
 | What should I do next       | `pm next`                                | `quickstart` |
@@ -111,3 +115,4 @@ Aggregate on a scalar field when a per-value count is what you want.
 | Triage and planning procedures     | [Workflows](references/WORKFLOWS.md)          | ~350 tok |
 | Prompt templates                   | [Prompts](references/PROMPTS.md)              | ~250 tok |
 | Backlog structure and item quality | [Backlog shaping](references/BACKLOG_SHAPING.md) | ~900 tok |
+| Complete ecosystem review and long-term planning | [Ecosystem planning](references/ECOSYSTEM_PLANNING.md) | on demand |

package/.agents/skills/pm-user/references/ECOSYSTEM_PLANNING.md

@@ -0,0 +1,95 @@
+# Complete Ecosystem Planning
+
+Tracked by [pm-doxj](../../../pm/milestones/pm-doxj.toon),
+[pm-qnl2](../../../pm/tasks/pm-qnl2.toon), and
+[pm-e9zh](../../../pm/chores/pm-e9zh.toon).
+
+Use this procedure for an explicitly exhaustive portfolio review. Normal
+implementation starts with bounded context and expands only its active lineage.
+
+## Establish the Population
+
+```bash
+pm context --limit 10 --for orient
+pm search "<request keywords>" --limit 10
+pm list --status open --limit 20
+pm list --status in_progress --limit 20
+pm list --all --no-truncate --full --include-body --strict-read --json
+pm events --full --limit 1000
+```
+
+Stream large results to an analysis consumer instead of loading them into the
+agent's context. Certify the list's total, completeness, truncation, and omission
+receipt. For events, parse the terminal `pm.stream.trailer`, retain every event's
+complete `entry`, and resume with `--since <next_cursor>` until `has_more` is
+false. Record the snapshot cutoff: a concurrent workspace can change during a
+read, and new events belong to a subsequent delta. Event collection is not an
+independent hash-chain verification.
+
+Read `pm get <ID> --full --json` and the item's history before changing its
+meaning or making a status claim. A closed implementation remains completed
+when its encompassing outcome still has work left. Read the full contract only
+when enumerating the surface: `pm contracts --full --json`; its default summary
+does not contain the complete flag, action, SDK, and MCP catalog.
+
+## Inspect Meaning and Coverage
+
+```bash
+pm graph audit --full --json
+pm graph analyze --full --json
+pm duplicates --status all --limit 100
+pm validate --check-resolution --check-history-drift
+pm health --check-only --full
+```
+
+Inspect continuation receipts on every collection. Fuzzy duplicate clusters
+are review candidates: separate scanner identities, follow-ups, and parent/child
+deliveries can share titles or issue numbers without being duplicates.
+
+Check active acceptance criteria, readiness, risk, confidence, estimates,
+outcome reachability, and ownership. Compare actual repository files, docs,
+tests, packages, and command contracts with linked artifacts. Historical empty
+fields need evidence recovery, not invented estimates or generic completion
+text. Reuse a substantive closure statement or original acceptance criterion
+only with an explicit source comment, and keep unrecoverable gaps recorded.
+
+## Build a Useful Graph
+
+Use `implements` for delivery, `verifies` for evidence, `discovered_from` for
+provenance, `recurs_from` for the same failure mechanism, `supersedes` for
+replacement, and one ordering direction for a real prerequisite. Check the
+runtime relationship contract before choosing a kind.
+
+An identifier mentioned in a boundary statement does not establish a
+dependency. Neither file overlap nor timestamps alone prove causality. Do not
+add inverse duplicates, hierarchy chains, or generic links to achieve a depth
+quota. Preserve evidence references and uncertain candidates separately from
+accepted edges. Verify graph and history invariants after each bounded batch.
+
+## Plan Outcomes and Experiments
+
+Reuse the living milestone and canonical feature owners. Organize Current,
+Emerging, Expansion, and Exploratory horizons by evidence and prerequisite
+readiness; dates are forecasts with assumptions, not completion claims.
+
+Every independently testable proposal needs an existing owner or a deduplicated
+item with scope, non-goals, acceptance criteria, dependencies, expected value,
+risks, and an evaluation. Algorithm proposals also name data inputs, bounded
+candidate generation, time/memory cost, deterministic fallback, provenance,
+negative controls, and a measurable promotion/rejection rule. Keep historical
+catalog aliases resolvable; use stable slugs for new concepts.
+
+Use the actual workflow capabilities that help: plans for execution steps,
+atomic `pm item mutate` batches for heterogeneous changes, notes for proposals,
+learnings for conclusions, and assurance measurements for regression gates.
+Read each unfamiliar command's help first. Record adopt/cover/waive decisions
+for capabilities that would otherwise require artificial production data.
+
+For this repository's algorithm and cognitive proposals, consult the
+[portfolio index](../../../../docs/CONTEXT_ALGORITHM_PORTFOLIO.md), then read
+the individual owner. Historical ordinal references require their source-note
+timestamp; a closed substrate does not prove its proposed extensions shipped.
+
+After terminal metadata changes, regenerate and check the package-owned
+changelog, preserving historical release attribution. Release claims when the
+batch ends; keep unfinished programmes open with exact residual scope.

package/.claude-plugin/marketplace.json

@@ -6,14 +6,14 @@
   },
   "metadata": {
     "description": "Official marketplace for pm CLI — native git-based project management for Claude Code and AI coding agents.",
-    "version": "2026.9.6"
+    "version": "2026.9.8"
   },
   "plugins": [
     {
       "name": "pm-claude",
       "source": "./plugins/pm-claude",
       "description": "Native pm CLI integration for Claude Code — 28 MCP tools, 5 workflow skills, 14 slash commands, 4 subagents, hybrid TUI task tracking, session context injection, and coordination subagents for git-based project management without leaving Claude Code.",
-      "version": "2026.9.6",
+      "version": "2026.9.8",
       "author": {
         "name": "unbrained",
         "url": "https://github.com/unbraind/pm-cli"