project-loop-harness 0.4.1__tar.gz → 0.4.3__tar.gz

{project_loop_harness-0.4.1/src/project_loop_harness.egg-info → project_loop_harness-0.4.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: project-loop-harness
-Version: 0.4.1
+Version: 0.4.3
 Summary: Local control plane for agentic project development loops
 Author: Project Loop Harness Contributors
 License: MIT
@@ -245,6 +245,85 @@ Markdown.
 See [docs/context-pack.md](docs/context-pack.md) for the contract shape and
 boundaries.
 
+## Work Briefs
+
+An optional `work-brief/v1` artifact can capture a target's reviewed intent,
+acceptance criteria, constraints, non-goals, and assumptions without changing
+the default `pcl start` path:
+
+```bash
+pcl contract validate --type work-brief/v1 work-brief.json --json
+pcl brief add work-brief.json --summary "Reviewed execution input" --dry-run --json
+pcl brief add work-brief.json --summary "Reviewed execution input" --json
+pcl brief review E-0001 --actor "agent:codex" --actor-kind agent \
+  --reason "Self-review completed; human gate remains open" --json
+# Normally run by the agent after the human says "approve" in conversation:
+pcl brief approve E-0001 --actor "human:owner" --actor-kind human \
+  --recorded-by "agent:codex" --recorder-kind agent \
+  --source-kind conversation --source-ref "conversation:<approval-reference>" \
+  --reason "Human explicitly approved the presented review packet" --json
+pcl brief show --target task:T-0001 --json
+pcl route recommend --target task:T-0001 --json
+pcl policy explain --target task:T-0001
+pcl route override --target task:T-0001 --profile assure \
+  --actor "human:owner" --reason "Require independent review" --dry-run --json
+pcl route override --target task:T-0001 --profile assure \
+  --actor "human:owner" --reason "Require independent review" --json
+pcl route current --target task:T-0001 --json
+```
+
+Brief content is immutable Evidence. Review and approval are separate
+hash-bound events: agent/system review cannot satisfy the human approval gate.
+Task context, resume, and dashboard data expose the factual actor kind, actor,
+recorder, source reference, timestamp, target, and bound hash without treating
+assumptions as facts. Humans normally approve in conversation or Cockpit; the
+agent records that decision through PCL. Direct human CLI use remains a
+compatibility path, not the expected routine UX. See
+[docs/work-brief-v1.md](docs/work-brief-v1.md).
+Deterministic Direct/Discover/Assure recommendation is documented in
+[docs/route-recommendation-v1.md](docs/route-recommendation-v1.md).
+Multi-axis resolution and field-level source rules are documented in
+[docs/adaptive-policy-v1.md](docs/adaptive-policy-v1.md).
+An applied override preserves its original recommendation and policy
+resolution as separate hash-bound Evidence. Task context, completion packets,
+and resume handoffs expose additive references without rewriting historical
+artifacts.
+
+## Evidence Sets
+
+Use `evidence-set/v1` when one passing artifact is not enough and the target
+must retain which declared reports were included or excluded:
+
+```bash
+pcl evidence-set plan --target task:T-0001 --work-root work/lp \
+  --manifest work/lp/reports/report-manifest.json \
+  --required-kind visual_check \
+  --include visual_check=E-0001:acceptance --json
+pcl evidence-set record --target task:T-0001 --work-root work/lp \
+  --manifest work/lp/reports/report-manifest.json \
+  --required-kind visual_check \
+  --include visual_check=E-0001:acceptance \
+  --summary "LP verification evidence set" --json
+```
+
+Planning is read-only. Recording creates immutable, target-linked Evidence and
+keeps known exclusions visible. See
+[docs/evidence-set-v1.md](docs/evidence-set-v1.md).
+
+Use a domain-neutral completion policy when a Test must require an external
+JSON verdict rather than merely retain the report set:
+
+```bash
+pcl completion evaluate --policy completion-policy.json \
+  --evidence-set E-0003 --test TC-0001 --json
+pcl test pass TC-0001 --summary "Completion contract passed" \
+  --evidence-id E-0003 --completion-policy completion-policy.json --json
+```
+
+Only allowlisted JSON predicates are evaluated. The Evidence Set must target
+the exact Test, remain complete, and retain matching report hashes. See
+[docs/completion-policy-v1.md](docs/completion-policy-v1.md).
+
 ## Explainable Code Context
 
 Build a local code context snapshot when an agent handoff needs auditable code

{project_loop_harness-0.4.1 → project_loop_harness-0.4.3}/README.md

@@ -211,6 +211,85 @@ Markdown.
 See [docs/context-pack.md](docs/context-pack.md) for the contract shape and
 boundaries.
 
+## Work Briefs
+
+An optional `work-brief/v1` artifact can capture a target's reviewed intent,
+acceptance criteria, constraints, non-goals, and assumptions without changing
+the default `pcl start` path:
+
+```bash
+pcl contract validate --type work-brief/v1 work-brief.json --json
+pcl brief add work-brief.json --summary "Reviewed execution input" --dry-run --json
+pcl brief add work-brief.json --summary "Reviewed execution input" --json
+pcl brief review E-0001 --actor "agent:codex" --actor-kind agent \
+  --reason "Self-review completed; human gate remains open" --json
+# Normally run by the agent after the human says "approve" in conversation:
+pcl brief approve E-0001 --actor "human:owner" --actor-kind human \
+  --recorded-by "agent:codex" --recorder-kind agent \
+  --source-kind conversation --source-ref "conversation:<approval-reference>" \
+  --reason "Human explicitly approved the presented review packet" --json
+pcl brief show --target task:T-0001 --json
+pcl route recommend --target task:T-0001 --json
+pcl policy explain --target task:T-0001
+pcl route override --target task:T-0001 --profile assure \
+  --actor "human:owner" --reason "Require independent review" --dry-run --json
+pcl route override --target task:T-0001 --profile assure \
+  --actor "human:owner" --reason "Require independent review" --json
+pcl route current --target task:T-0001 --json
+```
+
+Brief content is immutable Evidence. Review and approval are separate
+hash-bound events: agent/system review cannot satisfy the human approval gate.
+Task context, resume, and dashboard data expose the factual actor kind, actor,
+recorder, source reference, timestamp, target, and bound hash without treating
+assumptions as facts. Humans normally approve in conversation or Cockpit; the
+agent records that decision through PCL. Direct human CLI use remains a
+compatibility path, not the expected routine UX. See
+[docs/work-brief-v1.md](docs/work-brief-v1.md).
+Deterministic Direct/Discover/Assure recommendation is documented in
+[docs/route-recommendation-v1.md](docs/route-recommendation-v1.md).
+Multi-axis resolution and field-level source rules are documented in
+[docs/adaptive-policy-v1.md](docs/adaptive-policy-v1.md).
+An applied override preserves its original recommendation and policy
+resolution as separate hash-bound Evidence. Task context, completion packets,
+and resume handoffs expose additive references without rewriting historical
+artifacts.
+
+## Evidence Sets
+
+Use `evidence-set/v1` when one passing artifact is not enough and the target
+must retain which declared reports were included or excluded:
+
+```bash
+pcl evidence-set plan --target task:T-0001 --work-root work/lp \
+  --manifest work/lp/reports/report-manifest.json \
+  --required-kind visual_check \
+  --include visual_check=E-0001:acceptance --json
+pcl evidence-set record --target task:T-0001 --work-root work/lp \
+  --manifest work/lp/reports/report-manifest.json \
+  --required-kind visual_check \
+  --include visual_check=E-0001:acceptance \
+  --summary "LP verification evidence set" --json
+```
+
+Planning is read-only. Recording creates immutable, target-linked Evidence and
+keeps known exclusions visible. See
+[docs/evidence-set-v1.md](docs/evidence-set-v1.md).
+
+Use a domain-neutral completion policy when a Test must require an external
+JSON verdict rather than merely retain the report set:
+
+```bash
+pcl completion evaluate --policy completion-policy.json \
+  --evidence-set E-0003 --test TC-0001 --json
+pcl test pass TC-0001 --summary "Completion contract passed" \
+  --evidence-id E-0003 --completion-policy completion-policy.json --json
+```
+
+Only allowlisted JSON predicates are evaluated. The Evidence Set must target
+the exact Test, remain complete, and retain matching report hashes. See
+[docs/completion-policy-v1.md](docs/completion-policy-v1.md).
+
 ## Explainable Code Context
 
 Build a local code context snapshot when an agent handoff needs auditable code

project_loop_harness-0.4.3/agent-tasks/0146-work-brief-evidence-contract.md

@@ -0,0 +1,77 @@
+# 0146: Immutable Work Brief Evidence contract
+
+- **Status:** Done; human-approved 2026-07-11
+- **Milestone:** v0.4.2 Adaptive Entry
+- **Priority:** P0
+- **Estimated size:** M
+- **Dependencies:** v0.4.1 released; schema 8 consistent
+- **Parallel-safe with:** none
+- **DB schema:** remains 8
+
+## Problem
+
+Goal and Task state do not carry a compact, versioned execution input covering
+problem, outcome, acceptance criteria, constraints, non-goals, and assumptions.
+The integrated proposal embeds route output in the brief and stores mutable
+approval status inside an otherwise immutable artifact, creating a dependency
+cycle and unclear authority.
+
+## Goal
+
+Package `work-brief/v1` as immutable Evidence content. Link it to an existing
+target through generic Evidence links and record approval as a separate,
+hash-bound mutation event.
+
+## Contract
+
+- Required content: contract version, brief ID/revision, target, intent,
+  acceptance criteria, constraints, non-goals, assumptions, created-at/by.
+- Route is not required or embedded. A later optional
+  `route_recommendation_evidence_id` may reference 0147 output.
+- Artifact content never changes in place.
+- `pcl brief approve` records actor, reason, Evidence ID, target, and artifact
+  SHA-256 through the normal event/outbox transaction.
+- The authoritative current brief is the uniquely approved, non-revoked brief
+  for the target. Ambiguity fails closed.
+
+## Scope
+
+- Package JSON schema, validator, positive/negative fixtures, and CLI contract
+  validation.
+- Add read-only `pcl brief show` / target resolution.
+- Add explicit `pcl brief add` and `pcl brief approve` mutations reusing
+  Evidence storage/link services.
+- Add optional brief reference/summary to context pack and handoff packet.
+- Keep bare `pcl start` unchanged; any brief creation integration is explicit.
+
+## Invariants
+
+- No dedicated brief/intent table or migration.
+- Unapproved brief is never the critical execution contract.
+- Approval does not prove assumptions or Evidence claims.
+- Invalid schema, target mismatch, hash drift, or ambiguous approvals fail
+  before state/event/file mutation.
+- Read-only commands do not record usage or write artifacts by default.
+
+## Acceptance criteria
+
+- Valid brief can be copied as Evidence and target-linked.
+- Invalid brief creates no DB row, event, outbox record, or final artifact.
+- Approval is auditable and bound to the exact artifact hash.
+- Multiple active approvals fail with a typed ambiguity error.
+- Context/resume show the approved brief reference without inlining the full
+  body.
+- Projects without a brief retain byte-compatible default start/finish/resume
+  behavior.
+- Schema is present in wheel and sdist.
+
+## Required tests
+
+Schema fixtures; add/link/approve/show; hash/target/ambiguity negatives;
+read-only zero-mutation checks; packet/context compatibility; package-data and
+clean-wheel smoke.
+
+## Non-goals
+
+Route resolution, policy axes, revision/supersession workflow, LLM authoring,
+Markdown/YAML authoring, Option generation, or Knowledge.

project_loop_harness-0.4.3/agent-tasks/0147-deterministic-route-recommendation.md

@@ -0,0 +1,53 @@
+# 0147: Deterministic route recommendation
+
+- **Status:** Done; human-approved 2026-07-11
+- **Milestone:** v0.4.2 Adaptive Entry
+- **Priority:** P0
+- **Estimated size:** L
+- **Dependencies:** 0146
+- **Parallel-safe with:** none
+- **DB schema:** remains 8
+
+## Goal
+
+Produce a read-only `route-recommendation/v1` from explicit, normalized inputs
+without an LLM. The UX profile is `direct`, `discover`, or `assure`; reason
+codes and risk facts remain visible.
+
+## Contract
+
+Required fields: contract version, policy version, target, input digest,
+profile, risk level, normalized signals, and stable reason codes. Policy axes
+and override state are not part of this contract.
+
+## Scope
+
+- Pure ordered-rule resolver with documented tie breakers.
+- Signal catalog for acceptance completeness, ambiguity, requested paths,
+  dependency/migration/auth/security surfaces, deterministic checks, and
+  unknown data.
+- Read-only `pcl route recommend --target ... --json`.
+- Optional explicit `--record` stores the exact recommendation as Evidence and
+  appends one event/outbox record.
+- Additive start output only when the caller explicitly requests routing.
+
+## Invariants
+
+- Same normalized inputs and policy version yield byte-equivalent decision
+  content.
+- Model self-assessment is not a risk-lowering signal.
+- Missing information is explicit and cannot silently become Direct.
+- High-risk signals cannot be erased by lower-risk signals.
+- Read-only default does not mutate DB, events, outbox, or filesystem.
+
+## Acceptance criteria
+
+Fixtures prove clear low-risk -> Direct, missing acceptance/unverified cause ->
+Discover, and auth/migration/destructive scope -> Assure. Linux/Windows path
+normalization, tie-break ordering, unknown input, JSON stdout purity,
+determinism, explicit record, and old start snapshots are covered.
+
+## Non-goals
+
+Policy axes, enforcement, override, model selection, semantic code analysis,
+or automatic workflow launch.

project_loop_harness-0.4.3/agent-tasks/0148-adaptive-policy-resolve-explain.md

@@ -0,0 +1,52 @@
+# 0148: Adaptive policy resolve and explain
+
+- **Status:** Done; human-approved 2026-07-11
+- **Milestone:** v0.4.2 Adaptive Entry
+- **Priority:** P0
+- **Estimated size:** L
+- **Dependencies:** 0147
+- **Parallel-safe with:** none
+- **DB schema:** remains 8
+
+## Goal
+
+Resolve versioned multi-axis policy from a route recommendation and explain
+the source rule for every field without changing state.
+
+## Contract
+
+`adaptive-policy-resolution/v1` contains policy version/hash, recommendation
+Evidence/reference, input digest, resolved axes, per-field source rules,
+conflicts, and reason codes. Policy source is strict JSON (`adaptive-policy/v1`)
+parsed with the standard library.
+
+Axes are planning depth, verification depth, execution chunk size, checkpoint
+frequency, context budget, optional tool/time budgets, and escalation budget.
+
+## Scope
+
+- JSON policy schema and packaged default policy.
+- Deterministic precedence: defaults -> project rules -> risk floor.
+- Read-only `pcl policy resolve` and `pcl policy explain` JSON/text surfaces.
+- Typed errors for unknown keys, invalid values, conflicting same-precedence
+  rules, and unreadable policy.
+- Per-axis provenance suitable for packet references.
+
+## Invariants
+
+- Invalid policy never silently falls back to defaults.
+- Risk floors cannot be lowered by capability or budget signals.
+- Policy change does not reinterpret historical recorded resolutions.
+- No PyYAML or other runtime dependency.
+- Resolve/explain are read-only.
+
+## Acceptance criteria
+
+Rule precedence matrix, conflict/unknown-key negatives, deterministic output,
+policy hash stability, field-level explanation, JSON stdout purity, and Direct
+overhead measurements pass.
+
+## Non-goals
+
+Override mutation, enforcement, provider pricing, model/tool execution, or
+profile plugins.

project_loop_harness-0.4.3/agent-tasks/0149-route-policy-override-integration.md

@@ -0,0 +1,47 @@
+# 0149: Audited route/policy override and packet integration
+
+- **Status:** Done; human-approved 2026-07-11
+- **Milestone:** v0.4.2 Adaptive Entry
+- **Priority:** P0
+- **Estimated size:** L
+- **Dependencies:** 0148
+- **Parallel-safe with:** none
+- **DB schema:** remains 8
+
+## Goal
+
+Record an explicit operator override while preserving the original
+recommendation and policy resolution, then expose optional references through
+context, completion, and handoff packets.
+
+## Scope
+
+- `pcl route override --target --profile --reason --actor` with dry-run/preview.
+- Hash-bound references to separate original recommendation and resolution Evidence.
+- One transactional domain mutation, event, and outbox record.
+- Read-only current/effective route and override explanation.
+- Optional additive packet/context fields and backward-compatible fixtures.
+
+The accepted implementation records three Evidence rows and links in one
+transaction: original recommendation, original resolution, and the override.
+Only one aggregate `route_override_recorded` event/outbox pair is appended.
+
+## Invariants
+
+- Actor and non-empty reason are required.
+- Override never deletes or rewrites original artifacts.
+- Permission, destructive-operation, migration, and human-review floors remain
+  non-overridable.
+- Failure before commit leaves no DB/event/outbox/artifact trace.
+- Old packet readers/fixtures remain valid.
+
+## Acceptance criteria
+
+Preview is zero-mutation; successful override is audited; forbidden downgrade
+fails closed; original/effective views remain available; policy changes do not
+mutate historical packets; context/finish/resume compatibility suites pass.
+
+## Non-goals
+
+Automatic override, enforcement of all axes, Replan, stale propagation,
+Discovery Profile, or automatic model selection.

project_loop_harness-0.4.3/agent-tasks/0149a-adaptive-entry-dogfood.md

@@ -0,0 +1,19 @@
+# 0149a: Adaptive Entry dogfood and release gate
+
+- **Status:** Done; human-reviewed 2026-07-11
+- **Milestone:** v0.4.2 Adaptive Entry
+- **Priority:** P0
+- **Dependencies:** 0146–0149
+- **DB schema:** remains 8
+
+Exercise PLH itself and at least one external repository with clear,
+ambiguous, and high-risk tasks. Record recommendation, policy explanation,
+override decision, outcome, confusion, and local timing as project-local
+Evidence. Direct route must add no mandatory human step; resolver p95 target is
+below 50 ms on the deterministic fixture batch.
+
+The gate also compares audit output to
+`docs/canonical-state-baseline-v0.4.2.md`: no new anomaly is allowed. Full
+pytest, ruff, Python matrix, Windows smoke, package-data, sdist/wheel, and clean
+install checks are required. A human reviews dogfood conclusions; this task
+does not self-approve route quality.

project_loop_harness-0.4.3/agent-tasks/0149b-v0.4.2-release-preparation.md

@@ -0,0 +1,12 @@
+# 0149b: v0.4.2 local release preparation
+
+- **Status:** Done; local RC prepared, not published
+- **Milestone:** v0.4.2 Adaptive Entry
+- **Priority:** P0
+- **Dependencies:** 0149a human-reviewed dogfood
+- **DB schema:** remains 8
+
+Prepare version surfaces, release notes, fresh sdist/wheel, contract-package
+verification, clean-wheel smoke, and artifact hashes. This task authorizes
+local release preparation only. Tagging, pushing, GitHub Release creation, and
+PyPI publication require the release operator's explicit publication action.

project_loop_harness-0.4.3/agent-tasks/0150-evidence-set-completeness-contract.md

@@ -0,0 +1,49 @@
+# 0150: Evidence-set completeness contract
+
+- **Status:** Done locally; implemented and verified, not committed
+- **Milestone:** v0.4.3 Evidence Completeness
+- **Priority:** P0
+- **Dependencies:** v0.4.2 local RC; approved `docs/plan-v0.4.3.md`
+- **DB schema:** remain 8; stop for human approval if a migration is needed
+
+## Problem
+
+Hash-pinned Evidence proves what was selected, not that the selection is a
+complete account of the target. LP dogfood selected passing reports while a
+known 35.3% coordinate report and missing completion verdict remained outside
+the Evidence row.
+
+## Scope
+
+- Define and schema-validate a deterministic `evidence-set/v1` artifact.
+- Bind it to one target, an explicit work root, and a declared report manifest.
+- Record included Evidence, required report kinds, known exclusions, and a
+  completeness assessment.
+- Provide read-only plan/inspect output before record mutation.
+- Warn when a bundle excludes discovered related reports; reject terminal use
+  only when target policy marks the report required or blocking.
+- Package the schema and freeze canonical JSON/text fixtures.
+
+## Invariants
+
+- No arbitrary repository or parent-directory scan.
+- Filesystem proximity alone never creates an authoritative relationship.
+- Discovery order and finding order are deterministic.
+- Existing Evidence rows remain immutable and usable.
+- A rejected command leaves zero Evidence, link, or event traces.
+
+## Non-scope
+
+- Domain-specific verdict semantics.
+- Test/Feature terminal mutations.
+- Approval provenance.
+- DB migration, hosted indexes, or LLM classification.
+
+## Acceptance criteria
+
+1. Schema, validator, package-data tests, and canonical fixtures exist.
+2. The 35.3%-excluded fixture reports an incomplete required evidence set.
+3. Optional unrelated sibling files do not block completeness.
+4. Path escape, symlink escape, malformed JSON, missing manifest, and duplicate
+   role cases fail closed with zero mutation traces.
+5. JSON/text output is deterministic under reversed unordered SQLite selects.

project_loop_harness-0.4.3/agent-tasks/0151-completion-policy-terminal-preflight.md

@@ -0,0 +1,51 @@
+# 0151: Completion-policy adapter and terminal preflight
+
+- **Status:** Done locally; verified; not committed
+- **Milestone:** v0.4.3 Evidence Completeness
+- **Priority:** P0
+- **Dependencies:** 0150
+- **DB schema:** remain 8; stop for human approval if a migration is needed
+
+## Problem
+
+PCL can validate durable proof without knowing whether a collaborating tool
+classified the result as `prototype` or `complete`. It also permits Test
+planning without a Story and only discovers the structural weakness later.
+
+## Scope
+
+- Define a domain-neutral `completion-policy/v1` with allowlisted JSON
+  predicates; do not execute arbitrary expressions.
+- Bind the policy, evidence-set receipt, required verdict artifact, and
+  structured acceptance conditions to a Test preflight.
+- Require `complete` when target policy says so; permit honest intermediate
+  states when the acceptance contract only asks for a prototype.
+- Under enforced policy, reject `pcl test plan` without `--story` before
+  mutation. Preserve an advisory migration path for existing projects.
+- Make every terminal rejection transactionally clean.
+
+## Invariants
+
+- Core contains no `mockup-to-code`, DOM, or web-specific branches.
+- Policy evaluation is local, deterministic, JSON-only, and standard-library
+  first.
+- `--evidence-id` remains the canonical terminal-proof route.
+- A complete external verdict does not override unresolved required failures.
+- Existing valid direct Test transitions remain compatible.
+
+## Non-scope
+
+- Next-action priority changes.
+- Human/agent approval authority.
+- Third-party tool execution or report generation.
+
+## Acceptance criteria
+
+1. A `prototype` verdict is rejected when a Test requires `complete`.
+2. A missing required verdict and an incomplete evidence set are rejected with
+   zero Test, Feature, link, or event traces.
+3. A complete verdict plus required passing reports permits the existing
+   Evidence-ID-first terminal transition.
+4. Enforced and advisory Story-link planning behavior is fixture-tested.
+5. Schema, evaluator, inspect output, and packaged contracts are deterministic
+   and backward compatible.

project_loop_harness-0.4.3/agent-tasks/0152-next-action-approval-provenance.md

@@ -0,0 +1,52 @@
+# 0152: Next-action and approval-provenance integrity
+
+- **Status:** Done locally; verified; not committed
+- **Milestone:** v0.4.3 Evidence Completeness
+- **Priority:** P0
+- **Dependencies:** 0151
+- **DB schema:** remain 8; stop for human approval if a migration is needed
+
+## Problem
+
+LP dogfood reached a neutral idle route while its Feature was only `passing`
+and its domain verdict was incomplete. Agent self-review and explicit human
+approval also need visibly different authority.
+
+## Scope
+
+- Route `passing` but not `done` Features to a factual next action.
+- Route missing/incomplete required completion verdicts ahead of idle.
+- Add schema-8 event/Evidence receipts exposing `actor_kind`, factual actor
+  identity when available, source, timestamp, target, and bound hash.
+- Keep human-gated transitions unresolved unless provenance is human-origin.
+- Surface provenance and blockers in JSON, resume/context, and dashboard data
+  without turning the dashboard into state.
+- Treat conversation/Cockpit approval as the normal human UX: an agent may
+  record it later, but the receipt must distinguish approver from recorder and
+  bind the source reference. Humans are not expected to run routine CLI gates.
+
+## Invariants
+
+- `passing` and `done` remain distinct states.
+- The router does not invent readiness or approval.
+- Agent and system actions cannot claim `actor_kind=human`.
+- JSON additions are additive and deterministic.
+- Existing defect, decision, and escalation priority remains intact unless the
+  spec names the exact new comparison.
+
+## Non-scope
+
+- Identity federation or authentication.
+- Multi-user permissions.
+- Automatic Feature completion.
+
+## Acceptance criteria
+
+1. A passing unfinished Feature produces an actionable non-idle route.
+2. Missing required verdicts explain the artifact/policy blocker and safe next
+   command.
+3. Agent self-review cannot satisfy a human approval requirement.
+4. Explicit human approval produces a hash-bound provenance receipt.
+5. JSON/text/dashboard ordering and compatibility snapshots are deterministic.
+6. Agent-mediated recording of conversational approval preserves separate
+   human approver, agent recorder, and source reference fields.

project_loop_harness-0.4.3/agent-tasks/0153-cross-skill-integrity-dogfood.md

@@ -0,0 +1,51 @@
+# 0153: Cross-skill integrity dogfood and release gate
+
+- **Status:** Done locally; implementation, automated dogfood, and human review complete; not committed or released
+- **Milestone:** v0.4.3 Evidence Completeness
+- **Priority:** P0
+- **Dependencies:** 0150, 0151, 0152
+- **DB schema:** remains 8
+
+## Problem
+
+The contract is useful only if bundled Skill instructions and a real
+cross-skill workflow cannot reproduce the LP dogfood false-completion path.
+
+## Scope
+
+- Add canonical incomplete-prototype and complete-deliverable fixtures.
+- Re-run the LP-shaped flow or equivalent clean-room fixture with independent
+  human review.
+- Make all bundled `project-control-loop` Skill copies use parser-valid,
+  Evidence-ID-first terminal examples.
+- Explain that raw `--evidence` is a compatibility claim, not equivalent
+  terminal proof.
+- Verify exclusion warnings, policy mapping, unfinished routing, and approval
+  provenance end to end.
+- Prepare v0.4.3 release artifacts only after the human dogfood gate.
+
+## Invariants
+
+- Dogfood reports record failures and limitations, not only positive evidence.
+- Prototype and complete results remain distinct.
+- Package artifacts freeze only after full suite and clean-install smoke.
+- Tag, push, GitHub Release, and PyPI are separate explicit operations.
+
+## Non-scope
+
+- The external mockup skill's Motion Phase, crop generator, detail inventory,
+  or line-count repair.
+- Publishing a release.
+
+## Acceptance criteria
+
+1. The incomplete fixture cannot pass a completion-required Test and leaves
+   zero rejected-mutation traces.
+2. The complete fixture can pass with a reviewable evidence/provenance chain.
+3. `pcl next` never returns idle for the unfinished fixture.
+4. All bundled Skill examples are exercised against the real parser.
+5. Strict validation, full pytest, package/contract validation, and clean-wheel
+   smoke pass.
+6. Independent human review is given in conversation/Cockpit and recorded by
+   the agent with distinct human-approver, agent-recorder, and source fields
+   before release preparation; the human does not need to run `pcl`.

project_loop_harness-0.4.3/agent-tasks/0153b-v0.4.3-release-preparation.md

@@ -0,0 +1,16 @@
+# 0153b: v0.4.3 local release preparation
+
+- **Status:** Done; local release candidate prepared, publication excluded
+- **Milestone:** v0.4.3 Evidence Completeness
+- **Priority:** P0
+- **Dependencies:** 0153 human-reviewed dogfood
+- **DB schema:** remains 8
+
+Prepare the v0.4.3 version surfaces and release notes, run the canonical full
+suite and strict validation after the final tracked edit, build fresh sdist and
+wheel artifacts, verify package contracts and a clean-wheel installation, and
+record artifact hashes in a local release handoff.
+
+This task authorizes local release preparation and a local release commit only.
+Tagging, pushing, GitHub Release creation, and PyPI publication remain separate
+explicit release-operator actions.