synergyspec-selfevolving 1.4.0 → 2.0.0

package/README.md

@@ -133,7 +133,7 @@ Now tell your AI: `/synspec:propose <what-you-want-to-build>`
 → **[Workflows](docs/workflows.md)**: combos and patterns<br>
 → **[Commands](docs/commands.md)**: slash commands & skills<br>
 → **[CLI](docs/cli.md)**: terminal reference<br>
-→ **[Evaluation Lab](docs/evaluation-lab.md)**: advanced self-evolution harness<br>
+→ **[Self-Evolution Loop](docs/evaluation-lab.md)**: the loop-v2 in-context RL surface<br>
 → **[Supported Tools](docs/supported-tools.md)**: tool integrations & install paths<br>
 → **[Concepts](docs/concepts.md)**: how it all fits<br>
 → **[Multi-Language](docs/multi-language.md)**: multi-language support<br>
@@ -199,9 +199,9 @@ The self-evolution machinery lives in-tree under `src/core/fitness` and
 `src/core/self-evolution`, exposed through the `synergyspec-selfevolving
 self-evolution` subcommands (not as an ordinary day-to-day skill). It treats the
 spec workflow as a learnable system — a change is a forward pass through the
-artifact templates; `learn` runs the backward pass (a per-change loss); and a
-genetic-algorithm outer loop selects among competing candidate template
-variants.
+artifact templates; `learn` runs the backward pass (a per-change loss); and an
+in-context-RL episode improves one design 策略 POLICY lineage from a graded
+advantage, with a rollback that fires *before* any new edit.
 
 What actually works today:
 
@@ -233,21 +233,31 @@ What actually works today:
   session: an unresolvable flag is an up-front error (exit non-zero), while a
   missing env handle fails closed — no trajectory, and the observed-verified
   gate refuses to promote.
-- **Code-health gate** (auto-evolve / `evolve-from-edits`): a measured code-health
-  regression vs the last accepted state blocks auto-promotion (and surfaces a
-  loud `health-signal-unavailable` observation if a configured analyzer can't
-  run). No health signal ⇒ no gate, so the loop is never blocked on a missing
+- **Code-health gate** (the episode's evolving agent / `evolve-from-edits`): a
+  measured code-health regression vs the last accepted state blocks promotion (and
+  surfaces a loud `health-signal-unavailable` observation if a configured analyzer
+  can't run). No health signal ⇒ no gate, so the loop is never blocked on a missing
   measurement.
 - **Candidate proposals** (`self-evolution propose-canonical`): turns aggregated
   `learn` hints into human-gated candidate packages under
   `.synergyspec-selfevolving/self-evolution/candidates/`. Proposal-only — no
   canonical file is modified, and the frozen gen-test/run-test oracle is never
   touched.
-- **GA outer loop** (`self-evolution evolve`): groups candidates by canonical
-  target, scores them by accumulated fitness (or `--replay <corpus>` to re-run a
-  change corpus through baseline vs. candidate), ranks them, and generates a
-  **human-gated** promotion report. It never auto-promotes, and a frozen target
-  (per the per-target evolution policy) is skipped.
+- **In-context-RL episode** (`self-evolution episode --change <name>`): runs ONE
+  completed change through a fixed, code-spawned pipeline. A two-arm forward grades
+  the **主智能体 MAIN AGENT** (the frozen current 策略 POLICY) from the change's
+  `learn` report against a **CRITIC AGENT（基线智能体 baseline agent）** that reruns
+  the prior policy on the same change; a **奖励智能体 REWARD AGENT** computes
+  **advantage ＝ reward(主臂) − reward(基线臂)** and names the gap; on a bad advantage
+  the orchestrator rolls the 策略 POLICY back *before* the **演进智能体 EVOLVING
+  AGENT** is called; the evolving agent then makes ONE bounded edit (or refuses)
+  after the static / observed-GREEN / 范围⊆诊断 gates pass, writing the next policy
+  version onto your LOCAL files — no candidate dir, no republish. A frozen target
+  (per the per-target evolution policy) is skipped. Inspect the lineage with
+  `self-evolution policy show [--target <id>]`, re-enter a partial run with
+  `self-evolution episode resume <id>`, and manually restore a version with
+  `self-evolution policy rollback --target <id> --yes`. See
+  [docs/evaluation-lab.md](docs/evaluation-lab.md).
 - **Per-target evolution scope** (`selfEvolution:` in
   `synergyspec-selfevolving/config.yaml`): one switch decides which canonical
   targets may evolve, honored end-to-end by `learn` → `propose-canonical` → the
@@ -260,13 +270,16 @@ What actually works today:
 
 What is **not** built yet (don't infer it from the architecture diagrams):
 
-- No autonomous multi-generation **breeding** loop (mutate winners → next
-  generation). `evolve` runs a single generation over pre-existing candidates;
-  new candidates only come from the human-gated `propose-canonical` step.
+- No autonomous multi-episode **breeding** loop. Each `self-evolution episode`
+  improves the policy from one change's graded advantage; it never fans out
+  competing variants or runs unattended generations. New manual candidates only
+  come from the human-gated `propose-canonical` step.
 - No measured benchmark gains. The mechanisms are implemented and unit-tested,
   but the loop has not been run end-to-end to produce quantitative results.
-- The earlier out-of-tree `evolve/` DGM lab (and its `pnpm evolve` harness) has
-  been removed; the technique was internalized into `src/core`.
+- The earlier out-of-tree `evolve/` DGM lab (and its `pnpm evolve` harness) and
+  the genetic-algorithm outer loop (`self-evolution evolve` / `--replay`) have
+  been removed; the technique was internalized into `src/core` and re-homed on the
+  loop-v2 in-context-RL episode.
 
 ## Contributing
 

package/dist/commands/learn.d.ts

@@ -1,3 +1,14 @@
 import { Command } from 'commander';
-export declare function registerLearnCommand(program: Command): void;
+import { type RunEpisodeOptions, type RunEpisodeResult, type RunEpisodeBusy } from '../core/self-evolution/episode-orchestrator.js';
+/**
+ * Injectable dependencies for {@link registerLearnCommand}. The ONLY seam today
+ * is `runEpisode` — the loop-v2 episode runner the autonomous `--apply` entrance
+ * invokes. It defaults to the real {@link runEpisode}; learn tests stub it so the
+ * evolve path is exercised WITHOUT spawning real agents (the orchestrator's three
+ * agents each own a `runHeadlessAgent` spawn, which the stub never reaches).
+ */
+export interface LearnCommandDeps {
+    runEpisode?: (opts: RunEpisodeOptions) => Promise<RunEpisodeResult | RunEpisodeBusy>;
+}
+export declare function registerLearnCommand(program: Command, deps?: LearnCommandDeps): void;
 //# sourceMappingURL=learn.d.ts.map

package/dist/commands/learn.js

@@ -9,12 +9,14 @@ import { toTrajectoryFacts, describeRunnerResults } from '../core/trajectory/fac
 import { toActionSkeleton } from '../core/trajectory/skeleton.js';
 import { resolveHostHarness } from '../core/self-evolution/host-harness.js';
 import { mineSuccessSignals } from '../core/self-evolution/success-channel.js';
+import { captureMainArm, runEpisode, } from '../core/self-evolution/episode-orchestrator.js';
 import { buildLLMSummaryCandidates, ingestLearnHandoff, } from '../core/learn/llm-summary.js';
 function collect(value, previous) {
     previous.push(value);
     return previous;
 }
-export function registerLearnCommand(program) {
+export function registerLearnCommand(program, deps = {}) {
+    const runEpisodeImpl = deps.runEpisode ?? runEpisode;
     const learnCmd = program
         .command('learn [change]')
         .description('Review a completed change and extract reusable learning candidates')
@@ -157,13 +159,45 @@ export function registerLearnCommand(program) {
                     // side-write only; never fail learn over it.
                 }
             }
+            // LOOP-V2 AUTONOMOUS ENTRANCE (CS6-F): on an --apply run that opted into
+            // evolution, resolved EXACTLY ONE concrete evolvable target, and carries
+            // an observed-VERIFIED-GREEN signal, run one self-evolution episode
+            // in-process (rollback-before-evolution). A bare `learn <change>` preview
+            // can never reach this (it is neither --apply nor an evolving run), so the
+            // non-evolving path stays byte-identical to today. The runner is injected
+            // (`deps.runEpisode`) so tests exercise this path WITHOUT spawning real
+            // agents; it defaults to the real orchestrator.
+            let episodeOutcome;
+            const concreteEvolveTarget = resolveSingleConcreteTarget(evolutionPreview);
+            if (options.apply === true &&
+                isEvolvingRun(options) &&
+                concreteEvolveTarget !== undefined &&
+                concreteEvolveTarget.targetId !== null &&
+                reportIsObservedVerifiedGreen(report)) {
+                const mainArm = await captureMainArm({
+                    repoRoot: projectRoot,
+                    changeName: report.changeName,
+                    report,
+                });
+                episodeOutcome = await runEpisodeImpl({
+                    repoRoot: projectRoot,
+                    targetId: concreteEvolveTarget.targetId,
+                    changeName: report.changeName,
+                    changeDirPath: report.changeDir,
+                    mainArm,
+                });
+            }
             if (options.json) {
-                printJson(report, applied, evolutionPreview, hintsPath);
+                printJson(report, applied, evolutionPreview, hintsPath, episodeOutcome);
                 return;
             }
             console.log(renderLearnReport(report, applied));
             console.log('');
             console.log(renderLearnTransparency(report, applied, evolutionPreview, hintsPath, options));
+            if (episodeOutcome) {
+                console.log('');
+                console.log(renderEpisodeOutcome(episodeOutcome));
+            }
             if (successSummary) {
                 console.log('');
                 console.log(successSummary);
@@ -515,7 +549,7 @@ function renderIngestHandoff(changeName, ingest, applied) {
     }
     return lines.join('\n');
 }
-function printJson(report, applied, evolutionPreview, hintsPath) {
+function printJson(report, applied, evolutionPreview, hintsPath, episodeOutcome) {
     // `mode` only tracks whether MEMORY candidates were applied (--apply). It does
     // NOT reflect that --persist-hints wrote a hints file, which is what made the
     // old `mode:"preview"` read as "nothing written". `wrote` makes every write this
@@ -540,6 +574,30 @@ function printJson(report, applied, evolutionPreview, hintsPath) {
             hintIds: target.hintIds,
         })),
     };
+    // Loop-v2 episode outcome (CS6-F): present only when the autonomous entrance
+    // ran one this --apply run; absent on every non-evolving / unverified run so
+    // their --json stays byte-identical to today.
+    const episode = episodeOutcome
+        ? 'busy' in episodeOutcome
+            ? { busy: true, reason: episodeOutcome.reason }
+            : {
+                episodeId: episodeOutcome.episodeId,
+                baselineSkipped: episodeOutcome.baselineSkipped,
+                advantage: episodeOutcome.advantage,
+                decision: episodeOutcome.decision,
+                evolution: episodeOutcome.evolution === null
+                    ? null
+                    : episodeOutcome.evolution.kind === 'evolved'
+                        ? {
+                            kind: 'evolved',
+                            version: episodeOutcome.evolution.ledgerEntry.version,
+                        }
+                        : episodeOutcome.evolution.kind === 'refused'
+                            ? { kind: 'refused', reason: episodeOutcome.evolution.reason }
+                            : { kind: 'not-spawned', reason: episodeOutcome.evolution.reason },
+                newPolicyVersion: episodeOutcome.newPolicyVersion,
+            }
+        : undefined;
     console.log(JSON.stringify({
         mode: applied ? 'apply' : 'preview',
         ...report,
@@ -548,6 +606,7 @@ function printJson(report, applied, evolutionPreview, hintsPath) {
         ...(hintsPath ? { hintsPath } : {}),
         wrote,
         evolution,
+        ...(episode ? { episode } : {}),
     }, null, 2));
 }
 async function buildEvolutionPreview(hints, targetPolicy, projectRoot) {
@@ -715,10 +774,18 @@ function renderLearnTransparency(report, applied, evolutionPreview, hintsPath, o
     // full new content for that target's resolved LOCAL file.
     const concreteTargets = evolutionPreview.targets.filter((target) => target.targetId !== null);
     const concreteTarget = concreteTargets.length > 0 ? concreteTargets[0] : undefined;
-    if (hintsPath && concreteTarget) {
+    if (hintsPath && concreteTarget && concreteTarget.targetId !== null) {
+        const concreteTargetId = concreteTarget.targetId;
         lines.push(`- Hints written: ${hintsPath}`);
+        // LOOP-V2 (autonomous, rollback-before-evolution): the in-context-RL episode
+        // is now the default autonomous path. `--apply` on a verified-green run runs
+        // it in-process; this line is the explicit re-runnable form.
+        lines.push(renderEpisodeNextStep(report.changeName, concreteTargetId, options));
         const localFile = concreteTarget.localFiles[0] ?? concreteTarget.files[0] ?? '<target file>';
-        lines.push(`- Evolve from your edits: synergyspec-selfevolving self-evolution evolve-from-edits --from-learn "${hintsPath}" --evolve-target ${concreteTarget.targetId} --from-edits <edits.json> --yes`);
+        // MANUAL channel (host agent authors the edit directly): kept for operators
+        // who want to hand evolve-from-edits a full new-file content themselves
+        // rather than spawn the loop-v2 EVOLVING AGENT.
+        lines.push(`- Manual: evolve from your own edits: synergyspec-selfevolving self-evolution evolve-from-edits --from-learn "${hintsPath}" --evolve-target ${concreteTargetId} --from-edits <edits.json> --yes`);
         lines.push(`  you (the host agent) author edits.json's full new file content for the target file shown above (${localFile}).`);
     }
     else if (hintsPath) {
@@ -754,26 +821,99 @@ function renderLearnTransparency(report, applied, evolutionPreview, hintsPath, o
         lines.push('- After reviewing or evolving, run /synspec:archive to close the change.');
     }
     else {
+        // A bare preview has no agent-in-the-loop signal, so point the human/cron
+        // operator at the LOOP-V2 autonomous entrance (in-context RL,
+        // rollback-before-evolution): `self-evolution episode`.
         lines.push('');
-        lines.push('headless fallback (no host agent):');
-        lines.push(`- One-button local evolve: synergyspec-selfevolving self-evolution auto-evolve --change "${report.changeName}"${renderTargetArgs(options)}`);
+        lines.push('autonomous loop-v2 (rollback-before-evolution):');
+        lines.push(`- One-button local evolve: ${renderEpisodeCommand(report.changeName, undefined, options)}`);
         lines.push('- After reviewing or evolving, run /synspec:archive to close the change.');
     }
     return lines.join('\n');
 }
+/**
+ * The loop-v2 autonomous entrance command (CS6-F):
+ * `self-evolution episode --change "<name>" [--target <id>] [--session-id <id>]`.
+ * Replaces the GA `auto-evolve` / autonomous `evolve-from-edits` suggestions as
+ * the loop-v2 path. The `--target` pin is included only when a concrete target
+ * id resolved; `--session-id` is threaded through when the operator pinned an
+ * explicit trajectory handle so the episode grades the SAME session learn did.
+ */
+function renderEpisodeCommand(changeName, targetId, options) {
+    const parts = [`synergyspec-selfevolving self-evolution episode --change "${changeName}"`];
+    if (targetId)
+        parts.push(`--target ${targetId}`);
+    if (options.sessionId)
+        parts.push(`--session-id ${options.sessionId}`);
+    return parts.join(' ');
+}
+/** The loop-v2 next-step line shown when a concrete target is pinned. */
+function renderEpisodeNextStep(changeName, targetId, options) {
+    return `- Autonomous loop-v2 episode: ${renderEpisodeCommand(changeName, targetId, options)}`;
+}
 /**
  * An "evolving run" is one where the operator opted into evolution
  * (`--apply` / `--persist-hints` / a named `--evolve-target`) — per the skill,
  * the bare CLI previews and only the skill/agent flow passes these flags, so
  * this is the agent-in-the-loop proxy (the same signal that gates the
- * unbindable-hint observations in the learn action). The headless
- * `auto-evolve` fallback (it spawns its proposer internally) is for runs with
- * NO agent in the loop; the skill forbids the headless proposer when an agent
- * IS the proposer, so the fallback is suppressed on evolving runs.
+ * unbindable-hint observations in the learn action). The loop-v2 autonomous
+ * fallback (the `self-evolution episode` next-step, which code-spawns the
+ * reward + evolving agents) is for runs with NO agent in the loop; the skill
+ * forbids surfacing that autonomous entrance when an agent IS already the
+ * proposer, so the fallback is suppressed on evolving runs.
  */
 function isEvolvingRun(options) {
     return (options.apply === true || options.persistHints === true || options.evolveTarget !== undefined);
 }
+/**
+ * The report carries an observed-VERIFIED-GREEN signal: a REAL test runner was
+ * observed (`verified`) AND it came back green (`observedStatus === 'success'`).
+ * This is the SAME condition the failure-evidence routing uses in core/learn.ts
+ * (a verified-green run yields an empty failure list); the loop-v2 episode is
+ * only auto-launched when the main arm actually has a measured, trusted outcome
+ * — never on an authored-only or unverified report.
+ */
+function reportIsObservedVerifiedGreen(report) {
+    const facts = report.fitnessSample?.trajectoryFacts;
+    return facts !== undefined && facts.verified === true && facts.observedStatus === 'success';
+}
+/**
+ * The ONE concrete evolvable target this learn run resolved, or `undefined` when
+ * there is not exactly one (zero, or a kind-only/ambiguous group that still
+ * needs a `--evolve-target` pin). Mirrors the `concreteTarget` the next-steps
+ * renderer picks; the loop-v2 entrance refuses to guess when more than one
+ * concrete target is in play.
+ */
+function resolveSingleConcreteTarget(preview) {
+    const concrete = preview.targets.filter((target) => target.targetId !== null);
+    return concrete.length === 1 ? concrete[0] : undefined;
+}
+/** Render the loop-v2 episode outcome for the human-readable transparency block. */
+function renderEpisodeOutcome(episode) {
+    if ('busy' in episode) {
+        return `- Loop-v2 episode: skipped — another in-flight episode holds the target (${episode.reason}).`;
+    }
+    const lines = [];
+    lines.push(`- Loop-v2 episode: ${episode.episodeId}`);
+    const advantage = episode.advantage === null ? 'n/a (baseline skipped or abstained)' : String(episode.advantage);
+    lines.push(`  advantage: ${advantage}; decision: ${episode.decision}`);
+    if (episode.evolution === null) {
+        lines.push('  evolution: not spawned');
+    }
+    else if (episode.evolution.kind === 'evolved') {
+        lines.push(`  evolution: evolved -> policy v${episode.evolution.ledgerEntry.version}`);
+    }
+    else if (episode.evolution.kind === 'refused') {
+        lines.push(`  evolution: refused (${episode.evolution.reason})`);
+    }
+    else {
+        lines.push(`  evolution: not spawned (${episode.evolution.reason})`);
+    }
+    if (episode.newPolicyVersion !== null) {
+        lines.push(`  policy head: v${episode.newPolicyVersion}`);
+    }
+    return lines.join('\n');
+}
 function renderExplicitPolicy(explicit) {
     if (explicit.length === 0)
         return '';

package/dist/commands/self-evolution-episode.d.ts

@@ -0,0 +1,177 @@
+/**
+ * Loop-v2 CLI commands (self-evolution as in-context RL).
+ *
+ * These commands drive ONE bounded episode of the loop documented in
+ * `src/core/self-evolution/episode-orchestrator.ts`:
+ *
+ *   - `episode`               — run ONE episode for a change: build the 主智能体
+ *                               MAIN AGENT (frozen actor, 策略 POLICY vN+1) arm
+ *                               from a learn report (the same grading the `learn`
+ *                               command uses), then run the orchestrator (CRITIC
+ *                               AGENT（基线智能体 baseline agent）arm + 奖励智能体
+ *                               REWARD AGENT scoring + rollback/keep decision +
+ *                               演进智能体 EVOLVING AGENT optimizer.step). Prints a
+ *                               human summary (advantage, decision, evolution
+ *                               kind, new policy version) or --json.
+ *   - `episode resume <id>`   — re-enter a partially-run episode at its recorded
+ *                               stage and finish the remaining steps.
+ *   - `policy show`           — READ-ONLY: print the 版本账本 ledger (versions,
+ *                               actions, Δ stats, predictions) + the 否决缓冲
+ *                               reject-buffer for the target(s). Replaces the
+ *                               read-only role of the (soon-removed) `trajectory`
+ *                               command.
+ *   - `policy rollback`       — manual snapshot rollback to the prior version +
+ *                               a `human-reject` 否决缓冲 entry; requires --yes.
+ *
+ * Like the rest of the self-evolution CLI surface, every programmatic entrypoint
+ * here is exported and fully INJECTABLE (the orchestrator spawn seam, the learn
+ * report generator, stdout/stderr, clock) so the commands are unit-testable
+ * without spawning a real `claude` binary.
+ */
+import { Command } from 'commander';
+import { captureMainArm as captureMainArmImpl, resumeEpisode as resumeEpisodeImpl, type RunEpisodeOptions, type RunEpisodeResult, type RunEpisodeBusy, type ResumeEpisodeResult, type PolicyLedgerEntry, type RejectBufferEntry } from '../core/self-evolution/index.js';
+import { type LearnReport } from '../core/learn.js';
+/**
+ * The options the `episode` command forwards to the runEpisode seam. A superset
+ * of the orchestrator's {@link RunEpisodeOptions} carrying the `--no-baseline`
+ * request as `skipBaseline`. The base orchestrator ignores the extra key (the
+ * CRITIC AGENT arm is gated by its own ledger read); a custom seam may read it.
+ */
+export type EpisodeRunOptions = RunEpisodeOptions & {
+    skipBaseline?: boolean;
+};
+/**
+ * The injectable orchestrator seam. Accepts the {@link EpisodeRunOptions}
+ * superset (the real {@link runEpisodeImpl}, typed for {@link RunEpisodeOptions},
+ * satisfies this because it accepts a subset of the fields).
+ */
+export type EpisodeRunner = (opts: EpisodeRunOptions) => Promise<RunEpisodeResult | RunEpisodeBusy>;
+/**
+ * Attach the loop-v2 `episode` + `policy` subcommands to the parent
+ * `self-evolution` command. Called once from {@link registerSelfEvolutionCommand}.
+ */
+export declare function attachSelfEvolutionEpisodeCommands(parent: Command): void;
+export interface RunEpisodeCommandArgs {
+    /** The completed change to run the episode for. */
+    changeName?: string;
+    /** Canonical target id to evolve. Defaults to the design artifact-template. */
+    target?: string;
+    /** Skip the CRITIC AGENT（基线智能体 baseline agent）arm for this episode. */
+    noBaseline?: boolean;
+    /** Explicit transcript handle (Claude transcript store only). */
+    transcript?: string;
+    /** Explicit Claude session id handle. */
+    sessionId?: string;
+    json?: boolean;
+}
+export interface RunEpisodeCommandOptions {
+    repoRoot: string;
+    stdout?: (l: string) => void;
+    stderr?: (l: string) => void;
+    /**
+     * Test seam: produce the change's learn report (the 主智能体 MAIN AGENT arm's
+     * grading). Defaults to {@link generateLearnReport}; tests inject a graded stub
+     * so the episode runs without a real agent trajectory.
+     */
+    generateReport?: (changeName: string) => Promise<LearnReport>;
+    /**
+     * Test seam: build the {@link MainArmCapture} from a learn report. Defaults to
+     * the orchestrator's {@link captureMainArmImpl}.
+     */
+    captureMainArm?: typeof captureMainArmImpl;
+    /**
+     * Test seam: the orchestrator that runs the three agents. Defaults to the real
+     * {@link runEpisodeImpl}; tests inject a fake so NO agents are spawned.
+     */
+    runEpisode?: EpisodeRunner;
+}
+export interface RunEpisodeCommandResult {
+    exitCode: number;
+    /** Present when the episode ran (not busy / not an error). */
+    result?: RunEpisodeResult;
+    /** Present when the target's in-flight slot was already held. */
+    busy?: RunEpisodeBusy;
+    error?: string;
+}
+/**
+ * Programmatic entrypoint for `self-evolution episode`. Exported so tests can
+ * drive the full episode flow with an injected orchestrator seam (no real agent
+ * spawn).
+ */
+export declare function runEpisodeCommand(args: RunEpisodeCommandArgs, opts: RunEpisodeCommandOptions): Promise<RunEpisodeCommandResult>;
+export interface RunResumeEpisodeCommandArgs {
+    episodeId: string;
+    json?: boolean;
+}
+export interface RunResumeEpisodeCommandResult {
+    exitCode: number;
+    result?: ResumeEpisodeResult;
+    error?: string;
+}
+/**
+ * Programmatic entrypoint for `self-evolution episode resume <id>`. Exported for
+ * tests; the orchestrator seam is injectable.
+ */
+export declare function runResumeEpisodeCommand(args: RunResumeEpisodeCommandArgs, opts: {
+    repoRoot: string;
+    stdout?: (l: string) => void;
+    stderr?: (l: string) => void;
+    resumeEpisode?: typeof resumeEpisodeImpl;
+}): Promise<RunResumeEpisodeCommandResult>;
+export interface RunPolicyShowCommandArgs {
+    /** Restrict to a single canonical target id. */
+    target?: string;
+    json?: boolean;
+}
+export interface PolicyShowTargetView {
+    targetId: string;
+    /** Lineage head version, or null when the lineage has not been initialized. */
+    head: number | null;
+    ledger: PolicyLedgerEntry[];
+    rejectBuffer: RejectBufferEntry[];
+}
+export interface RunPolicyShowCommandResult {
+    exitCode: number;
+    targets: PolicyShowTargetView[];
+    error?: string;
+}
+/**
+ * Programmatic entrypoint for `self-evolution policy show`. READ-ONLY: reads the
+ * 版本账本 ledger + 否决缓冲 reject-buffer for the target(s) and renders them.
+ * Never mutates anything.
+ */
+export declare function runPolicyShowCommand(args: RunPolicyShowCommandArgs, opts: {
+    repoRoot: string;
+    stdout?: (l: string) => void;
+    stderr?: (l: string) => void;
+}): Promise<RunPolicyShowCommandResult>;
+export interface RunPolicyRollbackCommandArgs {
+    target: string;
+    /** Why the version is being rejected (recorded on the 否决缓冲 entry). */
+    reason?: string;
+    /** Required confirmation. */
+    yes?: boolean;
+    json?: boolean;
+}
+export interface RunPolicyRollbackCommandResult {
+    exitCode: number;
+    /** The rollback ledger entry, present on success. */
+    entry?: PolicyLedgerEntry;
+    /** The version the lineage was restored TO. */
+    toVersion?: number;
+    error?: string;
+}
+/**
+ * Programmatic entrypoint for `self-evolution policy rollback`. Manually rolls
+ * the 策略 POLICY lineage back to the previous version (recorded as a NEW
+ * monotonic head, git-revert style) and appends a `human-reject` 否决缓冲
+ * reject-buffer entry so the next 演进智能体 EVOLVING AGENT step sees the rejected
+ * direction. Requires --yes.
+ */
+export declare function runPolicyRollbackCommand(args: RunPolicyRollbackCommandArgs, opts: {
+    repoRoot: string;
+    stdout?: (l: string) => void;
+    stderr?: (l: string) => void;
+    now?: () => Date;
+}): Promise<RunPolicyRollbackCommandResult>;
+//# sourceMappingURL=self-evolution-episode.d.ts.map