@gobing-ai/spur 0.4.31 → 0.4.34

package/.claude-plugin/marketplace.json

@@ -7,7 +7,7 @@
     "plugins": [
         {
             "name": "sp",
-            "version": "0.4.31",
+            "version": "0.4.34",
             "source": "./plugins/sp"
         }
     ]

package/config/config.global.yaml

@@ -167,7 +167,7 @@ decisions:
     paths: []
     maker: typesafe
     # makers:
-    #     task-semantic-risk: laya-local
+    #     failure-class: laya-local
     # What "the decision was right" means (task 1263 R5): per decision id, the recorded outcomes
     # that count as correct for each served value. `String(value)` keys, so a yes/no decision uses
     # "true"/"false". The outcome verb records them (`spur decision outcome <invocation-id> ...`);
@@ -177,6 +177,3 @@ decisions:
         failure-class:
             fix: [fixed]
             stop: [not-fixed]
-        task-semantic-risk:
-            "false": [none, p4, p3]
-            "true": [p1, p2]

package/config/decisions/task-pipeline.yaml

@@ -2,7 +2,7 @@ $schema: "@gobing-ai/spur/schemas/decision-catalog.schema.json"
 # Decision catalogs for the task pipeline (feature P1, ADR-134).
 #
 # The shipped `config/workflows/task-pipeline.yaml` decide actions reference these ids
-# (`failure-class`, and `task-semantic-risk` once its evidence bar is met). Each entry's
+# (`failure-class`). Each entry's
 # reliability is measured from recorded `decision.*` events — `spur decision status
 # --reliability` — and an adoption slice only keeps a decision point while that evidence
 # shows it adds information the deterministic guards do not.
@@ -17,34 +17,6 @@ defaults:
     minConfidence: 0.8
 
 decisions:
-    task-semantic-risk:
-        type: noul
-        description: >
-            Does this task's change set change behaviour a reviewer must reason about beyond what
-            the diffstat already shows? (false projects the fast lane, true or a fallback the
-            standard lane.)
-        instructions: |
-            You are classifying one Spur task's change set for review depth.
-            Change set: ${params.instructions}
-            Answer true when the change alters behaviour a reviewer must reason about: control
-            flow, a data shape, persistence, security, an external contract, or error handling.
-            Answer false only when the change is mechanical — docs, comments, formatting, renames,
-            test-only edits, or a dependency bump with no source edit.
-            Judge the change set only; do not score its quality or estimate its size.
-        parameters:
-            wbs:
-                type: string
-                description: Task WBS whose change set is being classified.
-        criteria:
-            "true": >
-                The change set touches executable behaviour — a branch, a boundary, a stored
-                schema, a public signature, or a failure path — so a reviewer must read the diff,
-                not only its diffstat.
-            "false": >
-                The change set is mechanical — docs, comments, formatting, renames, test-only
-                edits, or a version bump with no executable change — so the diffstat is enough.
-        fallback: true
-
     failure-class:
         type: choice
         description: Is a failed quality gate fixable by the repair agent, or should the run stop?

package/config/rules/boundary/sp-runtime-path.yaml

@@ -24,8 +24,6 @@ exclude:
     # legitimately contains `config/...` documentation / self-references — not
     # runtime paths.
     - "apps/cli/config/**"
-    # Desktop stage copies the same generated build-time asset payload beside each binary.
-    - "apps/desktop/resources/**"
 rules:
     - id: sp-runtime-path
       description: >

package/config/rules/strict/runtime-boundaries.yaml

@@ -32,10 +32,6 @@ rules:
             - "**/tests/**"
             - "**/*.test.ts"
 
-            # Electron Node adapter: ts-runtime's streaming handle uses Bun.spawn.
-            # Owns the detached process-group handle needed for startup cancellation and quit.
-            - "apps/desktop/src/server-process.ts"
-
             # Low-level system primitives:
             # `spur builder pr-*` spine (script-absorption PR-7, moved from plugins/sp/scripts/pr-reviewing.ts):
             # every git/gh call goes through ONE synchronous, test-injectable CommandRunner seam
@@ -114,22 +110,16 @@ rules:
             - "packages/app/src/services/quality-gate.ts" # gate artifacts (log/findings/receipt) written directly; merged-fd gate log and fsync-free atomic writes must stay byte-stable (task 1006 R1; `spur task gate` since script-absorption PR-3; mirrors history-anatomy.ts)
             - "packages/app/src/workflow/wrapup/steps.ts" # `spur task wrapup` steps (script-absorption PR-5): sync status/metrics appends and run-artifact writes byte-pinned to the retired plugin scripts' golden output (apps/cli/tests/script-absorption-pr5-parity.test.ts); the FileSystem seam is async and has no append
             - "packages/app/src/workflow/wrapup/drift-probe.ts" # drift-probe step beside steps.ts (same exemption)
-            # Script-absorption PR-3: the residual sweep and triage diffstat glue moved here from the retired
-            # plugin scripts (`spur task residual` / `spur task diffstat`). Same class as quality-gate.ts:
+            # Script-absorption PR-3: the triage diffstat glue moved here from the retired
+            # plugin scripts (`spur task diffstat`). Same class as quality-gate.ts:
             # synchronous `wx`/tmp+rename artifact writes, mtime-ordered verdict copies, and an fstat(1)
             # dev/ino identity check (1067 R3) that the async FileSystem seam cannot express.
-            - "packages/app/src/services/residual-scan-run.ts"
             - "packages/app/src/workflow/task-diffstat.ts"
             - "packages/app/src/workflow/feature-verification-run.ts" # the pass opens the run's single merged log fd (`openSync(path,'a')`) and mkdirs the run dir beside the receipt writer, exactly as the plugin entry did before the move; the fd cannot be expressed through the FileSystem seam (task 1176 R2; same class as the quality-gate exemption above)
             - "packages/app/src/services/pr-review/index.ts" # `spur builder pr-*` (PR-7): sync --status-file verdict writes and the AGENTS.md review-rules probe inside the synchronous spine (same exemption as its process seam above)
             - "packages/app/src/services/run-storage.ts" # durable-evidence migration core (task 1025 R4): atomic tmp+rename byte copies with digest reread and atomic manifest publish must stay byte-stable (mirrors agent-usage-producer.ts / project-registry.ts exemptions)
             - "packages/app/src/services/durable-plane-ignore.ts" # .git/info/exclude append moved verbatim out of run-storage.ts to break an import cycle (same exemption)
 
-            # Electron Node adapters cannot load Bun runtime implementations; keep the
-            # synchronous existence/directory probes injected for unit testing (desktop contract).
-            - "apps/desktop/src/layout.ts"
-            - "apps/desktop/src/server-process.ts"
-
             # Synchronous bootstrap & path resolution:
             - "packages/config/src/bundled-config.ts" # sync bundledConfigRoot() directory walk
             - "packages/config/src/loader.ts" # sync resolveConfigFile() / schema specifier lookup

package/config/rules/structure/protected-files.yaml

@@ -37,9 +37,6 @@ rules:
             # this rule's include scope — scanning generated copies only adds noise
             # (e.g. `id: task-implement-transition` trips the sk- token shape).
             - "apps/cli/config/"
-            # Desktop ships the same generated config payload beside each compiled binary.
-            # Its source remains outside this source-code rule, just like the CLI bundle.
-            - "apps/desktop/resources/"
             # `apps/cli/plugins/` is the same kind of gitignored build:bundle output — a staged
             # npm copy of `plugins/`, itself outside the include scope. Minified generated libs
             # there tripped the api-key shape (0976; the generated libs were retired by script-absorption PR-6).

package/config/rules/surface/check-cli-surface.yaml

@@ -75,10 +75,9 @@ rules:
       # declare no --json of their own.
       - "apps/cli/src/commands/workflow/persist.ts"
       # Pipeline-internal task verbs (script-absorption PR-3) declare no --json: their contract is
-      # the .spur/run artifact they write (gate status/receipt, residuals.json, diffstat.json) and
+      # the .spur/run artifact they write (gate status/receipt, diffstat.json) and
       # the exit code; the gate's own status line is serialized by the app service.
       - "apps/cli/src/commands/task/gate.ts"
-      - "apps/cli/src/commands/task/residual.ts"
       - "apps/cli/src/commands/task/diffstat.ts"
       # Absorbed measurement/report verbs (script-absorption PR-6): the app module serializes the
       # golden-pinned JSON (step-profile-cli for `workflow profile --json`, run-summary-cli for

package/config/workflow-candidates.json

@@ -7,7 +7,7 @@
             "canonical": "task-pipeline",
             "deadline": "2026-11-24",
             "createdAt": "2026-09-25",
-            "rationale": "0943 (feature D64): task-pipeline work routes by triage lane and failure class instead of static graphs. Between a green gate and the review/verify fork a deterministic `triage` state writes the diffstat (`spur task diffstat`, formerly plugins/sp/scripts/task-diffstat.ts; numstat + untracked scan against the precheck-anchored run base), honors a caller-set mode verbatim (never overridden), pins the safety lane on sensitive paths (drizzle/, packages/config/, server auth basenames, *secret*, .github/, plugins/sp/hooks/, *.sql) or >400 changed lines with NO model call, otherwise consults `decide task-triage` (low → fast lane skips review; standard/high → review), and appends the route reason to .spur/memory/task-pipeline-routes.log (R5). On a red gate, `decide failure-class` routes retryable → test-recheck (attempt counted, qualityGateMaxFixAttempts still bounds), fix → test-fix, stop → failed(failed-check); missing/corrupt decision fails closed. Projected basis: the 0938 baseline (docs/reports/2026-09-workflow-cost-baseline.md) measured task-pipeline wall median 7983s over 1 real terminal run with agent.run median 0; the fast lane saves the review agent.run + wall on triage-low diffs, and retryable classification saves fixall hops (and their gate re-runs) on transient failures. Measurement gap honestly named: with workflow.decideDecisionMaker off (R4) triage degrades to standard lanes (diffstat-guard-sensitive/large still forced safety — pure shell, no model), so benefit only materializes with the flag on; per-lane route counts are observable post-merge in .spur/memory/task-pipeline-routes.log. Measured-benefit status: projected from 0938, NOT shadow-run-confirmed — plan step 5 (recorded-input shadow runs) is DEFERRED out of the 0943 session and tracked by this candidate's deadline; `promotion evaluate` runs post-merge. If it does not beat the baseline, it is reverted (ADR-076).",
+            "rationale": "0943 (feature D64): task-pipeline work routes by triage lane and failure class instead of static graphs. Between a green gate and the review/verify fork a deterministic `triage` state writes the diffstat (`spur task diffstat`, formerly plugins/sp/scripts/task-diffstat.ts; numstat + untracked scan against the precheck-anchored run base), honors a caller-set mode verbatim (never overridden), pins the safety lane on sensitive paths (drizzle/, packages/config/, server auth basenames, *secret*, .github/, plugins/sp/hooks/, *.sql) or >400 changed lines with NO model call, otherwise consults `decide task-triage` (low → fast lane skips review; standard/high → review), and appends the route reason to .spur/memory/task-pipeline-routes.log (R5). On a red gate, `decide failure-class` routes retryable → test-recheck (attempt counted, qualityGateMaxFixAttempts still bounds), fix → test-fix, stop → failed(failed-check); missing/corrupt decision fails closed. Projected basis: the 0938 baseline (docs/reports/2026-09-workflow-cost-baseline.md) measured task-pipeline wall median 7983s over 1 real terminal run with agent.run median 0; the fast lane saves the review agent.run + wall on triage-low diffs, and retryable classification saves fixall hops (and their gate re-runs) on transient failures. Measurement gap honestly named: with workflow.decideDecisionMaker off (R4) triage degrades to standard lanes (diffstat-guard-sensitive/large still forced safety — pure shell, no model), so benefit only materializes with the flag on; per-lane route counts are observable post-merge in .spur/memory/task-pipeline-routes.log. Measured-benefit status: projected from 0938, NOT shadow-run-confirmed — plan step 5 (recorded-input shadow runs) is DEFERRED out of the 0943 session and tracked by this candidate's deadline; `promotion evaluate` runs post-merge. If it does not beat the baseline, it is reverted (ADR-076). 2026-11-13: the lane no longer consults `decide task-triage` — `triage` is a deterministic diffstat rule (docs/tests-only → fast, sensitive or >400 lines → safety, else standard) after the maker proved to serve only its fallback; evaluate against that rule, not the retired decide.",
             "measurement": {
                 "workflow": "task-pipeline"
             },

package/config/workflows/idea-pipeline.yaml

@@ -213,7 +213,7 @@ states:
       - kind: agent.run
         options:
           agent: ${vars.planningAgent}
-          input: "Generate acceptance criteria for feature ${vars.featureId}. The operator's ask of record is .spur/run/${vars.__runId}-idea-input.md — idea persisted verbatim at start; authoritative. Read the feature file and author R-numbered BDD Gherkin scenarios per ac-style-guide.md. Tie every scenario to the `## Requirement inventory` of .spur/run/${vars.__runId}-idea-eval-report.md: under each `Scenario:` heading add a comment `# covers: I1, I3` listing covered ids — every non-`[deferred: ...]` item needs at least one covering scenario. Two `spur task check` rules bind AC text: (1) AC bullets copy scenario titles verbatim — the title is the byte-identical identity key of its bullet; (2) gate-language words (HITL, approval/approved, merged/merge event, content-gate, GATED, capstone standalone) are forbidden in titles, bodies, and enum values (L4.gate-language) — rephrase around them. Output only the Acceptance Criteria section body (gherkin fence included), no commentary, no heading."
+          input: "Generate acceptance criteria for feature ${vars.featureId}. The operator's ask of record is .spur/run/${vars.__runId}-idea-input.md — idea persisted verbatim at start; authoritative. Read the feature file and author R-numbered BDD Gherkin scenarios per ac-style-guide.md. Tie every scenario to the `## Requirement inventory` of .spur/run/${vars.__runId}-idea-eval-report.md: under each `Scenario:` heading add a comment `# covers: I1, I3` listing covered ids — every non-`[deferred: ...]` item needs at least one covering scenario. AC bullets copy scenario titles verbatim — the title is the byte-identical identity key of its bullet. Output only the Acceptance Criteria section body (gherkin fence included), no commentary, no heading."
           role: planner
           session: fresh
           answerFile: .spur/run/${vars.__runId}-idea-ac-content.md