ace-assign 0.58.0 → 0.61.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: c66fd1fc72e1bb186537931c960477dc7f9405ce3bc0f41c195a4b6601aa9bd1
-  data.tar.gz: fef52c4c68bbb1435499d7cb0468d09919b5b330d9d7226b4c676c54c12ff9ad
+  metadata.gz: 494e61c0842f8930425d12aeee6c0b51b46ce491e829c19401df1296e440ca09
+  data.tar.gz: 614e997a57ea09e7ff83d30ad372514dc0c0abceac7a809cdcb4aea1b630005a
 SHA512:
-  metadata.gz: 8e31f4379b900c7906be0b15b92768c904d57bf3fcfccbaf284ca972168413be443907281e1f1b939dff66337daf2835f32fc27c27fc802738dbe19238f220c4
-  data.tar.gz: f77a7e04992fddf9b8ddf99e57243759e1ec8ccbd7502fbae9836e32bd749a60b5feb7d7c4a63017f92b384efdce49654547b7ecb1d18fb29c66cab644dad18a
+  metadata.gz: a83405e18f47bdf33cf83619a88b790ad6ed27ff933f8816530931a2eb571c744f8eeca5c25aded6c0c2dacdda5429aec87372aadae63b0393a17a2797d149ea
+  data.tar.gz: c6dba5cc8f2925af84265864bbf4b54e3102f27d0707ee936d0440ec3df28e3acbf67cc7040a103c808461d33cfd77c8dd43f5092bbec6ce0fa362b1920cfc52

data/CHANGELOG.md

@@ -7,6 +7,44 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [0.61.0] - 2026-10-02
+
+### Added
+
+- Journal-backed service request claims, terminal receipts, and uncertainty
+  tracking. Assignment evidence now exposes unresolved service effects and
+  the latest evidence ref commit without changing the candidate head.
+
+### Fixed
+
+- Validate terminal service receipts against the exact claimed request and
+  executor evidence before accepting them in the assignment journal.
+
+### Technical
+
+- Updated the ace-tmux dependency to the 0.18 adapter release line.
+
+## [0.60.0] - 2026-10-01
+
+### Fixed
+- Preserve task/project attachment when the public assignment creation path archives its source configuration, so managed attempts retain immutable journal authority.
+- Cache-free evidence lookup handles a seed-only journal ref without creating a checkout or treating an absent attempt as a corrupt journal.
+
+### Added
+- `attempt evidence --kind review-approval` validates accepted independent approval separately from collection; retained collection/approval receipts support explicit historical-head validation without granting current check authority.
+- Bind campaign check names to accepted executed operations and review collection to accepted `review-collect` artifacts, through `attempt evidence` purpose checks.
+
+- `ace-assign attempt evidence` exposes read-only coordinator-accepted current-head check proof. Journal reads use immutable Git objects without writing cache/audit checkout state.
+- Review receipts may consume a durable ace-review campaign result. The existing coordinator still validates authority, live head, independent reviewer, executed checks and artifact digests; stale or conflicting campaign snapshots are rejected.
+
+## [0.59.0] - 2026-09-29
+
+### Added
+- Durable lifecycle exclusion (`Ace::Assign::Molecules::LifecycleExclusion`): prune holds the exclusive side from final evidence reads through removal while every supported start path -- attempt registration, fork/driver session launch, and worktree provisioning -- holds the shared side. Locks live outside every deletion target (sandbox-aware cache base or the shared Git common dir) so they survive removal of the target, and a `removed` marker makes post-prune starts fail closed instead of recreating a deleted identity.
+
+### Changed
+- `AttemptCoordinator#start` and `ForkSessionLauncher#launch` now participate in the prune exclusion: a concurrent prune blocks new writers for the identity, and a start targeting a pruned assignment raises a conflict instead of writing into a recreated path.
+
 ## [0.58.0] - 2026-09-28
 
 ### Added
@@ -503,7 +541,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Unified dynamic step default instructions under a single `AssignmentExecutor::DEFAULT_DYNAMIC_STEP_INSTRUCTIONS` source.
 - Limited canonical batch expansion to explicit `workflow`, `skill`, or declared `sub_steps` inputs, preserving prior numbering behavior for plain flat batch inserts.
 - Updated usage, getting-started, and handbook docs to cover the new `add --from` workflow.
-- Updated `record-demo` step instructions to require diagnosis before skipping on failure — agents must check config, available fonts, and spike findings before reporting a non-blocking skip.
+- Updated `record-demo` step instructions to require diagnosis before skipping on failure -- agents must check config, available fonts, and spike findings before reporting a non-blocking skip.
 - Added `decision_notes.non_blocking_policy` to `record-demo` step catalog entry clarifying that "non-blocking" means "diagnose and retry before skipping," not "skip on first failure."
 
 ### Technical
@@ -959,7 +997,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [0.20.0] - 2026-03-05
 
 ### Added
-- Provider-specific session detection fallback in `ForkSessionLauncher` — when a provider doesn't return a native `session_id`, scans local session storage via `SessionFinder` to detect the forked session by prompt matching.
+- Provider-specific session detection fallback in `ForkSessionLauncher` -- when a provider doesn't return a native `session_id`, scans local session storage via `SessionFinder` to detect the forked session by prompt matching.
 - Session metadata file (`<root>-session.yml`) written for every fork run, capturing `session_id`, `provider`, `model`, and `completed_at` for traceability.
 - Stall error messages now include `Session: <id>` when session metadata is available, enabling direct trace to agent session.
 
@@ -1044,14 +1082,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [0.16.1] - 2026-03-04
 
 ### Fixed
-- Correct `.agents/skills` typo to `.agent/skills` in default config and `SkillAssignSourceResolver` — skill discovery now uses the canonical provider-neutral path
+- Correct `.agents/skills` typo to `.agent/skills` in default config and `SkillAssignSourceResolver` -- skill discovery now uses the canonical provider-neutral path
 
 ## [0.16.0] - 2026-03-04
 
 ### Added
-- New `onboard-base` catalog step — loads base project context via `ace-bundle project-base`
-- New `task-load` catalog step — loads task behavioral spec via `ace-bundle task://<taskref>`
-- Taskref placeholder substitution in catalog step descriptions — `<taskref>` in step descriptions is replaced with actual task reference during child instruction building
+- New `onboard-base` catalog step -- loads base project context via `ace-bundle project-base`
+- New `task-load` catalog step -- loads task behavioral spec via `ace-bundle task://<taskref>`
+- Taskref placeholder substitution in catalog step descriptions -- `<taskref>` in step descriptions is replaced with actual task reference during child instruction building
 
 ### Changed
 - Default assignment presets updated to use ace-task
@@ -1065,7 +1103,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [0.15.1] - 2026-03-01
 
 ### Added
-- Fork-run crash recovery protocol in drive workflow — detection, commit partial work, progress report, inject recovery steps, re-fork pattern for partial completion scenarios
+- Fork-run crash recovery protocol in drive workflow -- detection, commit partial work, progress report, inject recovery steps, re-fork pattern for partial completion scenarios
 
 ## [0.15.0] - 2026-02-28
 
@@ -1222,7 +1260,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [0.12.14] - 2026-02-22
 
 ### Added
-- Subtree guard step in drive workflow — driver reviews all fork report files before continuing to next step
+- Subtree guard step in drive workflow -- driver reviews all fork report files before continuing to next step
 - Report review instruction in split-subtree-root step template for fork context
 
 ## [0.12.13] - 2026-02-22
@@ -1295,13 +1333,13 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [0.12.1] - 2026-02-19
 
 ### Fixed
-- `reflect-verify-cycle` pair changed from `sequential` to `conditional` — prevents overriding optional strength of reflect-and-refactor
+- `reflect-verify-cycle` pair changed from `sequential` to `conditional` -- prevents overriding optional strength of reflect-and-refactor
 - Renamed `max_recursion` to `max_reruns` in replan config for clarity
 
 ## [0.12.0] - 2026-02-19
 
 ### Added
-- New `reflect-and-refactor` step in catalog — analyzes implementation against ATOM principles and executes targeted refactoring before PR creation
+- New `reflect-and-refactor` step in catalog -- analyzes implementation against ATOM principles and executes targeted refactoring before PR creation
 - Composition rules for reflect-and-refactor: ordering (after verify, before mark-done/release/retro), pairs (verify cycle, fix cycle, replan cycle), and conditional suggestion
 - `create-retro` step now consumes `findings-report` from reflect-and-refactor as recommended prerequisite
 - `implement-with-pr` recipe updated to include optional reflect-and-refactor step after work-on-task
@@ -1315,7 +1353,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
-- `assignment_state` now checks `completed` before `failed` — assignments where all steps are done/failed correctly report `:completed` instead of `:failed`
+- `assignment_state` now checks `completed` before `failed` -- assignments where all steps are done/failed correctly report `:completed` instead of `:failed`
 
 ### Added
 
@@ -1906,7 +1944,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
-- CLI `start` command crashes with positional argument (`ace-assign start job.yaml`) because `option :config` requires `--config` flag — renamed command to `create` with `argument :config`
+- CLI `start` command crashes with positional argument (`ace-assign start job.yaml`) because `option :config` requires `--config` flag -- renamed command to `create` with `argument :config`
 - `ace-bundle wfi://prepare-assignment` fails due to missing project-level wfi:// protocol registration
 
 ### Added

data/docs/usage.md

@@ -107,6 +107,33 @@ ace-assign attempt reconcile --attempt ATTEMPT --receipt receipt.json
 
 `reserved -> running -> succeeded | failed | stopped | uncertain`; reconciliation may resolve `uncertain -> succeeded | failed`. Terminal attempts accept no new effects, and accepted history is append-only.
 
+### Read accepted execution authority
+
+`attempt evidence` reads a managed coordinator-accepted receipt from immutable journal history without transitions, cache writes, or audit checkout creation. Supply the accepted attempt ID and exact canonical receipt digest. The trusted local operator or OS-enforced service verifies the actual executed outcome before accepting its receipt; worker-authored success claims cannot accept themselves. This reader consumes that attestation and does not execute a test or model.
+
+```sh
+ace-assign attempt evidence --attempt ATTEMPT --receipt-digest SHA256 --format json \
+  --kind check --check-name tests
+ace-assign attempt evidence --attempt ATTEMPT --receipt-digest SHA256 --format json \
+  --kind review-collection
+ace-assign attempt evidence --attempt ATTEMPT --receipt-digest SHA256 --format json \
+  --kind review-collection --historical-head EXACT_RECORDED_HEAD
+ace-assign attempt evidence --attempt ATTEMPT --receipt-digest SHA256 --format json \
+  --kind review-approval
+```
+
+A check proof requires the live candidate head, intact accepted artifacts, and the matching executed operation: `tests` uses `test`, other explicit check names use that operation name. Review collection requires an accepted `review-collect` operation with a passed `review-execution` outcome and the retained metadata/report/prompt artifacts. Review approval requires an ordinary accepted `review` operation with the approved independent reviewer verdict, exact head and retained report artifacts; collection proof cannot substitute for approval. Both collection and approval support an exact `--historical-head`. Historical review validation exposes `historical: true` and the recorded head; it preserves accepted source authority across head drift and cannot certify live checks. Unknown receipts, changed artifacts, unaccepted/unmanaged history, wrong purposes and stale current queries fail closed. Retain the configured evidence ref alongside durable review campaigns.
+
+### Service effect claims
+
+External service operations (driven through `ace-lab service request`) claim their effect on the assignment journal before dispatch, so a retried or crashed request can never duplicate a side effect. The coordinator owns the trusted surface; callers never append journal events directly.
+
+- `Ace::Assign::Organisms::AttemptCoordinator#claim_service_request(binding)` records the claim and its attempt event atomically under a journal-wide request-ID index: the same request ID cannot be claimed twice, by another assignment, or against a stale candidate head. External-effect operations additionally require accepted review evidence at the current head, like `merge`.
+- `#service_request_status(request_id)` reads the authoritative claim state, including after local cache loss.
+- `#transition_service_request(request_id, state:, receipt:)` moves a claim to `succeeded` or `failed` only with a bound, non-secret receipt (exact request/attempt/project/operation/input digest/target fields, non-negative executor UID, and SHA-256 evidence references); rejected receipts never become approvals.
+- `#reject_service_request(binding, reason:)` durably records an auditable rejection without inventing a valid attempt.
+- Journal commits advance `journal_commit` only; they never change or authorize `candidate_head`, and `base_head` stays fixed at attempt start.
+
 ### Evidence storage and recovery modes
 
 - Task-attached (managed) attempts journal accepted evidence to the configured evidence Git ref (default `refs/ace/execution`) through an isolated, disposable audit checkout, outside the deliverable candidate branch. Journal commits never advance or exempt the reviewed candidate.
@@ -159,11 +186,14 @@ HITL stall behavior:
 - Canonical contract lives in `wfi://hitl` (`ace-hitl` package workflow).
 - If a step is failed with canonical message format `HITL: <id> <path>`, `ace-assign status` prints operator guidance with the matching `ace-hitl show <id>` command and available path hint.
 - Recommended resume flow:
+
   - `ace-hitl show <id>`
   - requester path (default): `ace-hitl wait <id>`
   - fallback path (when waiter inactive): `ace-hitl update <id> --answer "<decision>" --resume`
   - `ace-assign retry <failed-step> --assignment <assignment-id>`
+
 - Completion-attention flow:
+
   - When assignment work is complete but explicit user action is needed, create an approval HITL event (`kind=approval`) and include the resume instruction for `/as-assign-drive <assignment-id>`.
 
 ### `ace-assign step [STEP]`
@@ -267,6 +297,7 @@ Launch modes:
 - `headless`: force the existing provider subprocess path and never create tmux panes
 - `tmux`: require tmux context, create or reuse `<origin-window>-fs`, start a real interactive agent in a pane there via `ace-llm --interactive`, and send the scoped `/as-assign-drive <assignment>@<root>` handoff automatically. The fork window name uses the shared `ace-tmux` safe-name policy, so punctuation in the base window is replaced with `-`. Fork windows and panes are created detached, so the current tmux focus stays where the user left it.
 - `tmux`: require tmux context, create or reuse `<origin-window>-fs`, start a real interactive agent in a pane there via `ace-llm --interactive`, and send the scoped `/as-assign-drive <assignment>@<root>` handoff automatically
+
   - This mode consumes the shared `ace-tmux` runtime/control surface for tmux targeting, pane dispatch, and diagnostics.
   - The fork window name uses the shared `ace-tmux` safe-name policy, so punctuation in the base window is replaced with `-`.
   - Fork windows and panes are created detached, so the current tmux focus stays where the user left it.
@@ -338,6 +369,7 @@ When using preset-backed assignment creation (`ace-assign create --task ...`, `/
 - Terminal refs (`done`, `skipped`, `cancelled`) are skipped before queue expansion.
 - Mixed sets continue with remaining non-terminal refs and report skipped terminal refs.
 - If all requested refs are terminal, assignment creation stops with:
+
   - `All requested tasks are already terminal (done/skipped/cancelled): <refs>`
   - `No assignment created.`
 

data/lib/ace/assign/cli/commands/attempt/evidence.rb

@@ -0,0 +1,28 @@
+# frozen_string_literal: true
+require_relative "base"
+module Ace
+  module Assign
+    module CLI
+      module Commands
+        module Attempt
+          class Evidence < Ace::Support::Cli::Command
+            include Attempt::Base
+            desc "Read coordinator-accepted live execution or historical review authority"
+            option :attempt, required: true, desc: "Accepted execution attempt ID"
+            option :receipt_digest, required: true, desc: "Accepted receipt digest"
+            option :kind, default: "check", desc: "Evidence purpose: check, review-collection or review-approval"
+            option :check_name, default: "tests", desc: "Required check name (tests binds operation test)"
+            option :historical_head, desc: "Recorded collection/approval head for historical authority only"
+            option :format, default: "json", desc: "JSON evidence projection"
+
+            def call(**options)
+              raise Ace::Support::Cli::Error, "evidence supports --format json" unless options[:format] == "json"
+              emit_json(build_coordinator.evidence(attempt_id: options[:attempt], receipt_digest: options[:receipt_digest],
+                kind: options[:kind], check_name: options[:check_name], historical_head: options[:historical_head]))
+            end
+          end
+        end
+      end
+    end
+  end
+end

data/lib/ace/assign/cli.rb

@@ -31,6 +31,7 @@ require_relative "atoms/evidence_digest"
 require_relative "atoms/assignment_scope"
 
 # Molecules
+require_relative "molecules/lifecycle_exclusion"
 require_relative "molecules/assignment_manager"
 require_relative "molecules/attempt_store"
 require_relative "molecules/assignment_discoverer"
@@ -70,6 +71,7 @@ require_relative "cli/commands/attempt/start"
 require_relative "cli/commands/attempt/status"
 require_relative "cli/commands/attempt/finish"
 require_relative "cli/commands/attempt/reconcile"
+require_relative "cli/commands/attempt/evidence"
 
 module Ace
   module Assign
@@ -94,6 +96,7 @@ module Ace
         ["fork-run", "Run subtree in forked process"],
         ["attempt start", "Start a scoped attempt for an assignment step"],
         ["attempt status", "Show attempt status for an assignment"],
+        ["attempt evidence", "Read accepted current-head check evidence"],
         ["attempt finish", "Finish an attempt with a structured execution receipt"],
         ["attempt reconcile", "Reconcile an interrupted or uncertain attempt"]
       ].freeze
@@ -157,6 +160,7 @@ module Ace
       register "fork-session", wrap_command(Commands::ForkSession)
       register "attempt start", wrap_command(Commands::Attempt::Start)
       register "attempt status", wrap_command(Commands::Attempt::Status)
+      register "attempt evidence", wrap_command(Commands::Attempt::Evidence)
       register "attempt finish", wrap_command(Commands::Attempt::Finish)
       register "attempt reconcile", wrap_command(Commands::Attempt::Reconcile)
 

data/lib/ace/assign/models/evidence_event.rb

@@ -17,6 +17,7 @@ module Ace
         TYPES = %w[
           intent process_start receipt_accepted transition
           candidate_invalidated reconciliation
+          service_claim service_transition
         ].freeze
 
         # Build an event chained to a previous digest.

data/lib/ace/assign/models/execution_receipt.rb

@@ -24,7 +24,7 @@ module Ace
         ].freeze
 
         attr_reader :attempt_id, :assignment_id, :project_id, :scope, :operation,
-          :producer, :head, :verdict, :artifacts, :checks, :review, :recorded_at, :digest
+          :producer, :head, :verdict, :artifacts, :checks, :review, :campaign, :recorded_at, :digest
 
         # @param attempt_id [String] Attempt the receipt belongs to
         # @param assignment_id [String] Owning assignment ID
@@ -40,7 +40,7 @@ module Ace
         # @param recorded_at [Time] Receipt creation time
         # @param digest [String, nil] Canonical digest (computed when nil)
         def initialize(attempt_id:, assignment_id:, project_id:, scope:, operation:, producer:, head:, verdict:,
-          artifacts: [], checks: [], review: nil, recorded_at:, digest: nil)
+          artifacts: [], checks: [], review: nil, campaign: nil, recorded_at:, digest: nil)
           @attempt_id = attempt_id
           @assignment_id = assignment_id
           @project_id = project_id
@@ -52,6 +52,7 @@ module Ace
           @artifacts = artifacts
           @checks = checks
           @review = review
+          @campaign = campaign
           @recorded_at = recorded_at
           @digest = digest || Atoms::EvidenceDigest.digest(digest_payload)
         end
@@ -61,7 +62,7 @@ module Ace
         #
         # @return [Hash] Digest payload
         def digest_payload
-          {
+          payload = {
             "attempt_id" => attempt_id,
             "assignment_id" => assignment_id,
             "project_id" => project_id,
@@ -74,6 +75,8 @@ module Ace
             "checks" => checks,
             "review" => review
           }
+          payload["campaign"] = campaign if campaign
+          payload
         end
 
         # Convert to a hash for serialization.
@@ -102,6 +105,7 @@ module Ace
             artifacts: data["artifacts"] || [],
             checks: data["checks"] || [],
             review: data["review"],
+            campaign: data["campaign"],
             recorded_at: parse_time(data["recorded_at"]) || Time.now.utc,
             digest: data["digest"]
           )

data/lib/ace/assign/molecules/evidence_calculator.rb

@@ -39,6 +39,24 @@ module Ace
           review_receipt = receipt_currency(receipts, "review", head)
           release_receipt = receipt_currency(receipts, "release", head)
           unresolved_effects = collect_unresolved_effects(attempts)
+          if assignment&.managed?
+            requests = journal.service_requests(assignment.id)
+            # A rejected effect stays unresolved when its claim had been
+            # dispatched (authorization consumed): only attributable evidence
+            # that no effect occurred settles it, never the bare rejection.
+            unresolved_effects.concat(requests
+              .select { |request| unresolved_service_state?(request) }
+              .map { |request| "#{request["request_id"]}:#{request["operation"]}:#{request["state"]}" })
+            # Settled requests keep their authorization only while their
+            # receipt evidence remains intact in the candidate repository:
+            # a deleted or modified artifact drops the outcome back to
+            # unresolved instead of trusting an unverifiable receipt.
+            unresolved_effects.concat(requests
+              .select { |request| request["state"] == "succeeded" || request["state"] == "failed-settled" }
+              .reject { |request| evidence_intact?(request) }
+              .map { |request| "#{request["request_id"]}:#{request["operation"]}:evidence-unavailable" })
+            unresolved_effects.uniq!
+          end
           feedback_state = derive_feedback_state(attempts, receipts, head, unresolved_effects)
 
           # Authorization requires managed evidence, settled work (no active
@@ -67,7 +85,7 @@ module Ace
             base_head: latest&.binding&.base_head,
             candidate_head: latest&.candidate_head,
             evidence_git_ref: evidence_git_ref(assignment),
-            journal_commit: latest&.journal_commit,
+            journal_commit: assignment&.managed? ? journal.ref_value : latest&.journal_commit,
             unresolved_effects: unresolved_effects
           }
           evidence[:decision_digest] = Digest::SHA256.hexdigest(
@@ -78,6 +96,38 @@ module Ace
 
         private
 
+        # Accepted, uncertain, and failed effects are unresolved by
+        # definition; a rejected effect stays unresolved when its claim was
+        # dispatched (authorization consumed), because the effect may have
+        # happened.
+        def unresolved_service_state?(request)
+          %w[accepted uncertain failed].include?(request["state"]) ||
+            (request["state"] == "rejected" && request["consumed"] != false)
+        end
+
+        # Every receipt evidence artifact must still exist in the candidate
+        # repository with its recorded digest; evidence lives in the working
+        # tree, so its integrity is re-checked on every calculation.
+        def evidence_intact?(request)
+          receipt = request["receipt"]
+          return false unless receipt.is_a?(Hash)
+          repo_root = File.realpath(@repo_root)
+          Array(receipt["evidence"]).all? do |item|
+            path = File.expand_path(item["ref"].to_s, repo_root)
+            real = begin
+              File.realpath(path)
+            rescue Errno::ENOENT, Errno::EACCES, Errno::ELOOP
+              next false
+            end
+            begin
+              real.start_with?(repo_root + File::SEPARATOR) &&
+                Digest::SHA256.file(real).hexdigest == item["sha256"]
+            rescue Errno::EACCES, Errno::ELOOP
+              false
+            end
+          end
+        end
+
         def git_facts
           head, _s = Open3.capture2("git", "rev-parse", "HEAD", chdir: @repo_root, stdin_data: "")
           tree, _s = Open3.capture2("git", "rev-parse", "HEAD^{tree}", chdir: @repo_root, stdin_data: "")