synergyspec-selfevolving 2.1.5 → 2.1.7

package/dist/core/templates/workflows/learn.js

@@ -19,17 +19,19 @@ This is the review-and-learn step after \`/synspec:apply\` and \`/synspec:verify
    The runner starts with NO conversation context, so collect every handle it needs:
    - **Project root**: the absolute path of the current working directory.
    - **Change name**: from step 1.
-   - **Harness**: read the \`harness:\` key from \`synergyspec-selfevolving/changes/<name>/.synergyspec-selfevolving.yaml\`; if absent, use \`unknown\`.
-   - **Mode**: always \`apply\` — the episode runs the full loop (score, decide, and the 演进智能体's ONE bounded edit) autonomously, with no confirmation prompt. There is NO read-only episode and NO \`--preview\` flag. If the user wants a read-only look (no rollback, no evolution), do NOT run an episode: use the read-only view \`synergyspec-selfevolving self-evolution policy show\` (or a plain \`synergyspec-selfevolving learn <name>\` without \`--apply\`) instead.
-   - **Session handle (optional)**: if your harness exposes this session's id or transcript path, capture it; otherwise omit it (the 主智能体 MAIN AGENT arm's trajectory discovery then uses the change window).
+   - **Harness**: resolve the CURRENT host runtime, not the change metadata. If this skill is running in Codex, use \`codex\`; in Claude Code, use \`claude\`; in OpenCode, use \`opencode\`. Use \`unknown\` only when the host is genuinely unidentified after checking the active session/tooling. Do NOT read \`harness:\` from the per-change YAML for this field: that metadata is historical provenance, not the runtime that will spawn the loop-v2 agents.
+   - **Mode**: always \`apply\` — the episode runs the full loop (score, decide, and the 演进智能体's ONE bounded edit) autonomously, with no confirmation prompt. There is NO read-only episode and NO \`--preview\` flag. If the user wants a read-only look (no rollback, no evolution), do NOT run an episode: use the read-only view \`synergyspec-selfevolving self-evolution policy show\` (or a plain \`synergyspec-selfevolving learn <name>\` without \`--apply\`) instead.
+   - **Force-new episode**: \`yes\` only when the user explicitly asked to rerun / force a fresh episode; otherwise \`no\`. A normal learn run must not invent a rerun.
+   - **Isolation**: \`fresh-context subagent\` for the spawned runner.
+   - **Session handle (optional)**: if your harness exposes this session's id or transcript path, capture it; otherwise omit it (the 主智能体 MAIN AGENT arm's trajectory discovery then uses the change window).
 
 3. **Spawn the runner**
 
-   Use the host's available general-purpose Task/subagent runner (for example \`general-purpose\` on Claude or \`general\` on hosts that expose that type), prompt: "Use Skill tool to invoke synergyspec-selfevolving-self-evolving for change '<name>'. Project root: <root>. Harness: <harness>. Mode: apply. Session-id: <id>. Transcript: <path>. Trigger the loop-v2 self-evolution episode autonomously, do not ask the user questions, and end with the '## Episode Verdict' block."
+   Use the host's available general-purpose Task/subagent runner (for example \`general-purpose\` on Claude or \`general\` on hosts that expose that type), prompt: "Use Skill tool to invoke synergyspec-selfevolving-self-evolving for change '<name>'. Project root: <root>. Harness: <harness>. Mode: apply. Force-new: <yes|no>. Isolation: fresh-context subagent. Session-id: <id>. Transcript: <path>. Trigger the loop-v2 self-evolution episode autonomously, do not ask the user questions, and end with the '## Episode Verdict' block."
 
    Include the \`Session-id: <id>.\` / \`Transcript: <path>.\` segment only when the session handle from step 2 is known — omit it entirely when unknown.
 
-   The runner triggers exactly one CLI command — \`synergyspec-selfevolving self-evolution episode --change "<name>" --session-id <id>\` — and the orchestrator CODE-SPAWNS the 奖励智能体 REWARD AGENT + 演进智能体 EVOLVING AGENT (+ optional CRITIC AGENT（基线智能体）). Neither you nor the runner grades or edits canonical files.
+   The runner triggers exactly one CLI command — \`synergyspec-selfevolving self-evolution episode --change "<name>" --harness <harness> --session-id <id> --rerun\` when force-new is \`yes\`; omit \`--rerun\` when force-new is \`no\`; omit \`--harness\` when it is \`unknown\` — and the orchestrator CODE-SPAWNS the 奖励智能体 REWARD AGENT + 演进智能体 EVOLVING AGENT (+ optional CRITIC AGENT（基线智能体）). Neither you nor the runner grades or edits canonical files.
 
    Guardrails:
    - Do NOT trigger the episode yourself in this session — it must run from a fresh context.

package/dist/core/templates/workflows/run-tests.js

@@ -38,35 +38,53 @@ const INSTRUCTIONS_BODY = `**Input**: Optionally specify a change name. If omitt
      "startedAt": "<ISO timestamp>",
      "finishedAt": "<ISO timestamp>",
      "exitCode": 0,
-     "signal": null,
-     "stdoutLog": "synergyspec-selfevolving/changes/<name>/test-evidence/<timestamp>/runner.stdout.log",
-     "stderrLog": "synergyspec-selfevolving/changes/<name>/test-evidence/<timestamp>/runner.stderr.log",
-     "workspaceIdentity": {
-       "changeName": "<name>",
-       "taskId": "<benchmark task id, if any>",
-       "cwd": "<absolute working directory>",
-       "pyproject": {
-         "path": "pyproject.toml",
-         "name": "<[project].name, or null>",
-         "sha256": "<sha256 of pyproject.toml, or null>"
-       },
-       "packageJson": {
-         "path": "package.json",
-         "name": "<package.json name, or null>",
-         "sha256": "<sha256 of package.json, or null>"
-       }
-     },
-     "junitXml": null,
-     "coverageSummary": null,
-     "coverageLcov": null,
-     "coverageHtml": null
+     "signal": null,
+     "stdoutLog": "synergyspec-selfevolving/changes/<name>/test-evidence/<timestamp>/runner.stdout.log",
+     "stderrLog": "synergyspec-selfevolving/changes/<name>/test-evidence/<timestamp>/runner.stderr.log",
+     "stdoutLogSha256": "<sha256 of runner.stdout.log>",
+     "stderrLogSha256": "<sha256 of runner.stderr.log>",
+      "workspaceIdentity": {
+        "changeName": "<name>",
+        "taskId": "<benchmark task id, if any>",
+        "cwd": "<absolute working directory>",
+        "pyproject": null,
+        "packageJson": null
+      },
+     "testMetrics": {
+       "total": 29,
+       "passed": 29,
+       "failed": 0,
+       "passRate": 1
+     },
+     "junitXml": null,
+     "coverageSummary": null,
+     "coverageLcov": null,
+     "coverageHtml": null
    }
    \`\`\`
 
-   If the runner produces JUnit XML or coverage artifacts, record their paths in
-   \`runner-exit.json\`. If it does not, keep those fields \`null\`. The markdown
-   report may summarize results, but the raw logs and exit JSON are the durable
-   evidence that later verification must inspect.
+   Set each workspace identity file entry to an object ONLY when that file
+   exists at the project root. If \`pyproject.toml\` or \`package.json\` is absent,
+   leave that field \`null\` (or omit it); do not emit a \`path\` for an absent file.
+   Object shape for a present file:
+
+   \`\`\`json
+   {
+     "path": "pyproject.toml",
+     "name": "<project/package name, or null>",
+     "sha256": "<sha256 of the file>"
+   }
+   \`\`\`
+
+   If the runner summary exposes pass/fail counts, record them in
+   \`testMetrics\`; otherwise set \`testMetrics\` to \`null\` and preserve the raw
+   stdout/stderr logs. The \`stdoutLogSha256\` and \`stderrLogSha256\` fields MUST
+   be the SHA-256 hashes of the exact saved log files, computed after writing the
+   files and before writing \`runner-exit.json\`; do not hand-edit logs after
+   hashing. If the runner produces JUnit XML or coverage artifacts,
+   record their paths in \`runner-exit.json\`. If it does not, keep those fields
+   \`null\`. The markdown report may summarize results, but the raw logs and exit
+   JSON are the durable evidence that later verification must inspect.
 
 3b. **Promote PBT counterexamples to regression tests**
 
@@ -134,9 +152,10 @@ const INSTRUCTIONS_BODY = `**Input**: Optionally specify a change name. If omitt
    | UC1-E4a1 | Error when no grid space | ❌ failed | \`gridSize=0, widgetCount=1\` | \`test/pbt-regression-uc1-e4a1-1.test.ts\` |
    ...
 
-   ### Test Run Results
-   <summary from test runner output: passed/failed/skipped counts>
-   If failures: list failing test names and errors.
+   ### Test Run Results
+   Summary: <N collected>, <N passed>, <N failed>, <N skipped>, <N collection errors>
+   <raw summary from test runner output: passed/failed/skipped counts>
+   If failures: list failing test names and errors.
 
    ### Runner Evidence
    | Evidence | Path / Value |

package/dist/core/templates/workflows/self-evolving.js

@@ -12,9 +12,11 @@ You are the RUNNER for a completed SynergySpec-SelfEvolving change. In loop v2 (
 
 Parse these handles from the spawning prompt:
 - **Change name** (required). If the change name is missing or does not resolve via \`synergyspec-selfevolving list --json\`, stop and report the error — do NOT prompt the user (you may have no user channel).
-- **Absolute project root.** Run every CLI command from it.
-- **Harness**: \`claude\` | \`codex\` | \`opencode\` | \`unknown\`. If a harness was provided and differs from the ambient host, set \`SYNERGYSPEC_SELFEVOLVING_HOST_HARNESS=<harness>\` for the CLI invocation below.
-- **Session-id / transcript path** (optional). When the spawning prompt supplied a session-id or transcript path, pass \`--session-id <id>\` / \`--transcript <path>\` to the \`episode\` command so the 主智能体 MAIN AGENT arm's trajectory discovery does not depend on the change-window fallback.
+- **Absolute project root.** Run every CLI command from it.
+- **Harness**: \`claude\` | \`codex\` | \`opencode\` | \`unknown\`. If a concrete harness was provided, pass \`--harness <harness>\` to the CLI invocation below. If the prompt says \`unknown\` but this runner is clearly executing inside Codex, Claude Code, or OpenCode, recover the current host and pass that concrete harness. Omit \`--harness\` only when both the prompt and the current runner host are genuinely unidentified; never set \`SYNERGYSPEC_SELFEVOLVING_HOST_HARNESS=unknown\`.
+- **Force-new**: \`yes\` | \`no\` (optional; default \`no\`). If \`yes\`, append \`--rerun\` so a closed matching episode is not reused.
+- **Isolation**: \`fresh-context subagent\` | \`inline fallback (degraded)\` (optional). If supplied, copy it verbatim into the verdict; otherwise infer from whether this skill is running in a spawned subagent or inline fallback.
+- **Session-id / transcript path** (optional). When the spawning prompt supplied a session-id or transcript path, pass \`--session-id <id>\` / \`--transcript <path>\` to the \`episode\` command so the 主智能体 MAIN AGENT arm's trajectory discovery does not depend on the change-window fallback.
 
 **Recursion guard**
 
@@ -49,10 +51,11 @@ Everything in steps 1–6 is CODE. You do not perform any of it. You issue the c
 
    Run exactly ONE command — the loop-v2 orchestrator. It CODE-SPAWNS the 奖励智能体 REWARD AGENT + 演进智能体 EVOLVING AGENT (+ optional CRITIC AGENT（基线智能体）); you spawn nothing:
    \`\`\`bash
-   synergyspec-selfevolving self-evolution episode --change "<change>" --json
-   \`\`\`
-   - Append \`--session-id <id>\` and/or \`--transcript <path>\` ONLY when the spawning prompt supplied them.
-   - If the harness differs from the ambient host, set \`SYNERGYSPEC_SELFEVOLVING_HOST_HARNESS=<harness>\` first.
+   synergyspec-selfevolving self-evolution episode --change "<change>" --json
+   \`\`\`
+   - Append \`--session-id <id>\` and/or \`--transcript <path>\` ONLY when the spawning prompt supplied them.
+   - Append \`--harness <harness>\` when the spawning prompt supplied \`claude\`, \`codex\`, or \`opencode\`, or when the prompt supplied \`unknown\` but this runner can identify the current host as Codex, Claude Code, or OpenCode. Never append \`--harness unknown\`.
+   - Append \`--rerun\` ONLY when the spawning prompt supplied \`Force-new: yes\`.
 
    Do NOT grade, score, or author any edit yourself, and do NOT run \`evolve-from-edits\`, \`auto-evolve\`, or \`--agent\` / \`claude -p\` — those are not part of loop v2's host-facing path. The episode command IS the loop.
 
@@ -116,7 +119,7 @@ The session's final message MUST end with exactly this block shape:
 - Use \`busy-in-flight\` when the episode command returned the clean concurrency deferral (another in-flight episode holds the same 策略 POLICY target): advantage is null, episode id is none, 策略 POLICY version is unchanged. It is TRANSIENT and self-healing (retry after the lock clears / the 60-min stale window) — it is NOT a DEFECT, do not list it under Defects to surface, and never advise deleting \`in-flight.json\`.
 - When the episode did NOT start (Episode id is none — any not-run / busy-in-flight / error-* outcome), write \`none\` for Evolved target and Canonical file(s) changed, report Decision/Advantage as none/null, and leave 策略 POLICY version unchanged. The change's CONFIGURED target id is context only — do NOT copy it into the Evolved target field on a non-run verdict.
 - A \`kept\` / \`abstained\` outcome on a verified-green run is the CORRECT no-op, not a missed evolution — say so plainly rather than hedging.
-- Report \`Isolation: fresh-context subagent\` when you were spawned as a subagent; report \`Isolation: inline fallback (degraded)\` when this skill is running inline in the spawning session.`;
+- Copy the supplied \`Isolation:\` value verbatim when present. If it was not supplied, report \`Isolation: fresh-context subagent\` when you were spawned as a subagent, or \`Isolation: inline fallback (degraded)\` when this skill is running inline in the spawning session.`;
 export function getSelfEvolvingSkillTemplate() {
     return {
         name: 'synergyspec-selfevolving-self-evolving',

package/dist/core/trajectory/facts.d.ts

@@ -3,7 +3,7 @@ import type { HarnessName, NormalizedTrajectory } from './model.js';
 /** One failing test observed in the graded runner result's output. */
 export type ObservedTestFailure = ParsedTestFailure;
 export interface TrajectoryFacts {
-    harness: HarnessName;
+    harness: HarnessName | 'runner-evidence';
     changeName: string;
     /**
      * A recognizable test-runner invocation (vitest/pytest/go test/…) produced a

package/dist/core/trajectory/registry.js

@@ -5,12 +5,11 @@
  * Selection order:
  *   1. If the change metadata stamps a `harness` (the strongest signal — see
  *      `ChangeMetadata.harness`), use that adapter.
- *   2. Otherwise read the adapter for the harness that is ACTUALLY RUNNING this
- *      process (`resolveHostHarness()`: the CODEX_ / OPENCODE_ env heuristic, or
- *      the `SYNERGYSPEC_SELFEVOLVING_HOST_HARNESS` override). This makes the
- *      observed trajectory readable on a non-Claude host. With no harness env
- *      present it resolves to 'claude' — the historical default — so existing
- *      hermetic tests stay on the Claude adapter unchanged.
+ *   2. Otherwise read the trusted repo/host harness signal
+ *      (`resolveHostHarnessDetailsForRepo()`): explicit override, persisted
+ *      sidecar, explicit session id, or the historical Claude default. An
+ *      env-only Codex/OpenCode signal on an unstamped change is intentionally
+ *      too weak to scan the host's whole session store.
  *   3. `trajsz` is OPT-IN (env `SYNERGYSPEC_SELFEVOLVING_TRAJSZ`): when enabled
  *      and a fresh archive is present it is tried FIRST, since it already
  *      normalizes all three harnesses; absent/stale, we fall back to native.
@@ -98,8 +97,9 @@ export async function resolveTrajectorySource(projectRoot, changeName, options =
                     return s;
             }
             catch {
-                // fall through to probing
+                // fall through to the fail-closed return below
             }
+            return null;
         }
     }
     // 2. No stamp: use a trusted host recovery signal, not a blind global scan.
@@ -143,7 +143,7 @@ export async function getTrajectoryForChange(projectRoot, changeName, options =
 export async function getTrajectoryResultForChange(projectRoot, changeName, options = {}) {
     const source = await resolveTrajectorySource(projectRoot, changeName, options);
     if (!source)
-        return { trajectory: null, sourceHarness: null, reason: 'no-trajectory-source' };
+        return explainTrajectorySourceMiss(projectRoot, changeName, options);
     try {
         const result = source.getTrajectoryResult
             ? await source.getTrajectoryResult(changeName)
@@ -162,4 +162,35 @@ export async function getTrajectoryResultForChange(projectRoot, changeName, opti
         };
     }
 }
+async function explainTrajectorySourceMiss(projectRoot, changeName, options) {
+    const changeProvenance = await readChangeTrajectoryProvenance(projectRoot, changeName);
+    const sessionIds = uniqueNonBlank([
+        ...(options.sessionIds ?? []),
+        ...changeProvenance.sessionIds,
+        process.env.SYNERGYSPEC_SELFEVOLVING_SESSION_ID,
+    ]);
+    if (changeProvenance.harness) {
+        return {
+            trajectory: null,
+            sourceHarness: changeProvenance.harness,
+            reason: `stamped-${changeProvenance.harness}-source-unavailable`,
+        };
+    }
+    const hostResolution = await resolveHostHarnessDetailsForRepo(projectRoot);
+    const envOnlyNativeHost = hostResolution.source === 'env' &&
+        hostResolution.harness !== 'claude' &&
+        sessionIds.length === 0;
+    if (envOnlyNativeHost) {
+        return {
+            trajectory: null,
+            sourceHarness: hostResolution.harness,
+            reason: `env-only-${hostResolution.harness}-requires-stamped-harness-or-session-id`,
+        };
+    }
+    return {
+        trajectory: null,
+        sourceHarness: hostResolution.source === 'default' ? null : hostResolution.harness,
+        reason: `${hostResolution.source}-${hostResolution.harness}-source-unavailable`,
+    };
+}
 //# sourceMappingURL=registry.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "synergyspec-selfevolving",
-  "version": "2.1.5",
+  "version": "2.1.7",
   "description": "AI-native system for spec-driven development",
   "keywords": [
     "synergyspec-selfevolving",