dsh-doublecheck 0.6.0 → 0.7.1

package/lib/domain/projection.js

@@ -12,6 +12,7 @@
  */
 import { isTestCommand, isTestFilePath, joinTextBlocks, mutationTargetPath, parseRawArguments, shellCommand, testOutcome, } from "./evidence.js";
 import { SPEC_TOOL_NAME } from "./stages.js";
+import { countRedChecks } from "./gate.js";
 /** State for the empty log. */
 export function emptyDoublecheckState() {
     return {
@@ -21,6 +22,8 @@ export function emptyDoublecheckState() {
         specGoal: '',
         reviewed: false,
         editCount: 0,
+        gateVerdict: 'none',
+        gateRedCount: 0,
         pendingSpecCalls: {},
         pendingTestCalls: {},
     };
@@ -98,6 +101,13 @@ export function applyDoublecheckEvent(state, event, detection) {
             }
             return state;
         }
+        case 'doublecheck/gate': {
+            const gate = event.data;
+            const redCount = countRedChecks(Object.values(gate.phases));
+            if (state.gateVerdict === gate.verdict && state.gateRedCount === redCount)
+                return state;
+            return { ...state, gateVerdict: gate.verdict, gateRedCount: redCount };
+        }
         default:
             return state;
     }
@@ -128,5 +138,7 @@ export function viewDoublecheck(state) {
         specGoal: state.specGoal,
         reviewed: state.reviewed,
         editCount: state.editCount,
+        gateVerdict: state.gateVerdict,
+        gateRedCount: state.gateRedCount,
     };
 }

package/lib/domain/report.d.ts

@@ -14,8 +14,8 @@
 import type { SessionEvent } from '@deepseek-ai/dsh-session';
 import type { ObjectJsonSchema } from '@deepseek-ai/dsh-tools';
 import type { WorkflowMeta } from '@deepseek-ai/dsh-workflow';
-import { type TestRunDetection } from './evidence.ts';
-import { type GrilledSpec, type ReportVerdict, type ReviewFinding, type ReviewVerdict, type VerifyCheck } from './vocabulary.ts';
+import { type TestRunDetection } from './evidence.js';
+import { type GrilledSpec, type ReportVerdict, type ReviewFinding, type ReviewVerdict, type VerifyCheck } from './vocabulary.js';
 /** One timeline row of the report fold, in log order. */
 export interface ReportTimelineEntry {
     /** The evidence kind this row records. */

package/lib/domain/stages.d.ts

@@ -15,7 +15,7 @@
  * @module dsh-doublecheck/domain/stages
  */
 import type { SessionEvent } from '@deepseek-ai/dsh-session';
-import { type TestRunDetection } from './evidence.ts';
+import { type TestRunDetection } from './evidence.js';
 /** The discipline stages, in execution order. */
 export type DisciplineStage = 'grill' | 'design' | 'red' | 'green' | 'review' | 'verify';
 /** All discipline stages in order. */

package/lib/events.d.ts

@@ -14,14 +14,16 @@
  */
 import type { Agent } from '@deepseek-ai/dsh-agent';
 import type { Session, SessionEventMap } from '@deepseek-ai/dsh-session';
-import type { GrilledSpec, ReviewFinding, ReviewVerdict, ReportVerdict, VerifyCheck } from './domain/vocabulary.ts';
-export type { GrilledSpec, ReviewFinding, ReviewVerdict, ReportVerdict, VerifyCheck } from './domain/vocabulary.ts';
+import type { GrilledSpec, ReviewFinding, ReviewVerdict, ReportVerdict, VerifyCheck } from './domain/vocabulary.js';
+import type { GateState, GateVerdict } from './domain/gate.js';
+export type { GrilledSpec, ReviewFinding, ReviewVerdict, ReportVerdict, VerifyCheck } from './domain/vocabulary.js';
+export type { GateState } from './domain/gate.js';
 /** Guard enforcement strength. */
 export type GuardIntensity = 'remind' | 'warn' | 'block';
 /** Which discipline gate produced a guard reaction. */
-export type GuardGate = 'grill' | 'tdd' | 'delivery';
+export type GuardGate = 'grill' | 'tdd' | 'delivery' | 'gate';
 /** Policy outcome of one guard reaction. */
-export type GuardVerdict = 'reminded' | 'held' | 'denied' | 'green-pending' | 'report-expected';
+export type GuardVerdict = 'reminded' | 'held' | 'denied' | 'green-pending' | 'report-expected' | 'gate-red';
 declare module '@deepseek-ai/dsh-session' {
     interface SessionEventMap {
         /**
@@ -34,6 +36,14 @@ declare module '@deepseek-ai/dsh-session' {
         'doublecheck/state': {
             enabled: boolean;
         };
+        /**
+         * One settled delivery-gate run written by `/gate run`: the full gate
+         * state (checklist statuses, red items, verdict, engine, timestamp).
+         * Audit-safe by construction — counts, ids, and verdicts only — and the
+         * replayable source the `/gate status` panel and the turn-boundary red
+         * notice re-derive from. Rides `ignorable: true` like the state switch.
+         */
+        'doublecheck/gate': GateState;
     }
 }
 /**
@@ -46,6 +56,15 @@ export interface StateAppend {
         ignorable: true;
     }): void;
 }
+/**
+ * `Session.append` narrowed to the `doublecheck/gate` event, with the same
+ * `ignorable: true` envelope marker as the state switch.
+ */
+export interface GateAppend {
+    (type: 'doublecheck/gate', data: SessionEventMap['doublecheck/gate'], options: {
+        ignorable: true;
+    }): void;
+}
 declare module '@deepseek-ai/dsh-llm' {
     interface MessageSourceMap {
         /**
@@ -59,6 +78,17 @@ declare module '@deepseek-ai/dsh-llm' {
             verdict: ReviewVerdict;
             findings: ReviewFinding[];
         };
+        /**
+         * The delivery-gate red notice injected at the turn boundary: the
+         * model-facing text is the short gate notice (role statement first), and
+         * the structured verdict + red count ride this source so the durable
+         * once-semantics and the report folds read them without re-parsing prose.
+         */
+        'doublecheck-gate': {
+            kind: 'doublecheck-gate';
+            verdict: GateVerdict;
+            redCount: number;
+        };
     }
 }
 declare module '@deepseek-ai/cordis' {
@@ -151,5 +181,17 @@ declare module '@deepseek-ai/cordis' {
             path: string | null;
             written: boolean;
         }): void;
+        /**
+         * One delivery-gate run settled: the full gate state the `/gate run`
+         * command derived, after the durable `doublecheck/gate` session event was
+         * appended. Observability only — listeners must not veto or reroute.
+         * @param payload.session - the gated session.
+         * @param payload.state - the settled gate state (redacted by construction).
+         * @mode emit
+         */
+        'doublecheck/gate'(payload: {
+            session: Session;
+            state: GateState;
+        }): void;
     }
 }

package/lib/grill/index.d.ts

@@ -14,7 +14,7 @@
  */
 import type { Context } from '@deepseek-ai/cordis';
 import type Schema from '@deepseek-ai/schemastery';
-import type { GrilledSpec } from '../events.ts';
+import type { GrilledSpec } from '../events.js';
 export declare const name = "doublecheck-grill";
 export declare const inject: string[];
 /** Grill module configuration. */

package/lib/guard/command.d.ts

@@ -16,8 +16,8 @@
 import type { CommandInvocation, CommandResult } from '@deepseek-ai/dsh-commands';
 import { Session } from '@deepseek-ai/dsh-session';
 import type { SessionEvent } from '@deepseek-ai/dsh-session';
-import type { TestRunDetection } from '../domain/evidence.ts';
-import type { Config, Snapshot } from './index.ts';
+import type { TestRunDetection } from '../domain/evidence.js';
+import type { Config, Snapshot } from './index.js';
 export declare function hostStampsIgnorable(): boolean;
 /**
  * The session's doublecheck master switch: the last `doublecheck/state` event,

package/lib/guard/command.js

@@ -93,6 +93,9 @@ export function doublecheckHandler(deps) {
                     color: discipline.color,
                     reviewed: snapshot.lastReviewSeq >= 0,
                     editCount: discipline.editCount,
+                    gate: snapshot.lastGate === null
+                        ? null
+                        : { verdict: snapshot.lastGate.verdict, redCount: snapshot.lastGate.redCount },
                 }),
             };
         }

package/lib/guard/gate.d.ts

@@ -0,0 +1,170 @@
+/**
+ * The delivery gate runner: orchestrates the four-phase quality checklist
+ * over one session's durable evidence and settles the deliverable/rework
+ * decision.
+ *
+ * - **requirements** + **tests** — deterministic folds of the session log
+ *   (no model calls): the key-question checklist against the committed
+ *   spec, and the test-run/coverage evidence.
+ * - **consistency** — a local forked reviewer audits the diff ↔ requirement
+ *   mapping (every edit must serve a spec dimension).
+ * - **review** — the delivery conclusion. With `engine: auto`, the
+ *   dsh-auto-review engine's durable verdict records (`autoReview/verdict` /
+ *   `autoReview/rejection`) are the evidence; when the engine is not
+ *   installed or has no records this session, the phase degrades to the
+ *   local forked reviewer and says so. `engine: local` always uses the
+ *   local reviewer. The gate never synthesizes approval requests: that
+ *   chain may reach a human.
+ *
+ * Everything a run settles is audit-safe: counts, ids, verdicts, and
+ * redacted findings only. The settled state rides the durable
+ * `doublecheck/gate` session event (replay IS the state), the workspace
+ * `gate-report.md`, and the `/gate run` command result.
+ *
+ * @module dsh-doublecheck/guard/gate
+ */
+import type { Context } from '@deepseek-ai/cordis';
+import type { Agent } from '@deepseek-ai/dsh-agent';
+import type { CommandInvocation, CommandResult } from '@deepseek-ai/dsh-commands';
+import type { UserMessage } from '@deepseek-ai/dsh-llm';
+import type Schema from '@deepseek-ai/schemastery';
+import type { TestRunDetection } from '../domain/evidence.js';
+import { type GateQuestion, type GateState, type GateVerdict } from '../domain/gate.js';
+import type { GuardProse, ProseLanguage } from './prose.js';
+/** The reviewer knobs shared by the consistency and review phases. */
+export interface GateReviewerConfig {
+    enabled: boolean;
+    /** Subagent provider name the local reviewer runs on (default `fork`). */
+    provider: string;
+    /** Model route for the local reviewer; null means the main model reviews. */
+    model: string | null;
+    /** Tools the local reviewer may call (read-only by default). */
+    tools: string[];
+    /** Hard time budget for one local reviewer run. */
+    timeoutMs: number;
+    /** Maximum structured findings per phase. */
+    maxFindings: number;
+}
+/** The delivery-gate configuration block of the guard row. */
+export interface GateConfig {
+    /** Master switch for the gate panel and the turn-boundary red notice. */
+    enabled: boolean;
+    /** Suggest re-opening in plan mode when the verdict is rework. */
+    planSuggestion: boolean;
+    /** Workspace file (relative to the session cwd) receiving the gate report. */
+    reportFile: string;
+    /** The requirements-interrogation phase. */
+    requirements: {
+        enabled: boolean;
+        /** The configurable key-question checklist. */
+        checklist: GateQuestion[];
+        /** Minimum required questions that must be confirmed for the phase to pass. */
+        minConfirmed: number;
+        /** Tool name whose calls count as user interrogation evidence. */
+        interrogateTool: string;
+    };
+    /** The test-evidence phase. */
+    tests: {
+        enabled: boolean;
+        /** Require a passing test run as the latest evidence. */
+        requirePassingRun: boolean;
+        /** Failing runs after the latest green allowed before the gate turns red. */
+        allowFailingRuns: number;
+        /** Require coverage evidence in the test output. */
+        requireCoverage: boolean;
+        /** Minimum coverage percentage when coverage is required. */
+        minCoveragePct: number;
+        /** Regex (one capture group) parsing a coverage percentage from test output. */
+        coveragePattern: string;
+    };
+    /** The implementation-consistency phase (local forked reviewer). */
+    consistency: GateReviewerConfig;
+    /** The review-conclusion phase (engine verdict records + local degrade). */
+    review: GateReviewerConfig & {
+        /** `auto` = dsh-auto-review verdict records when present, else local. */
+        engine: 'auto' | 'local';
+    };
+}
+/** The default gate configuration: every phase on, the six key questions required. */
+export declare const DEFAULT_GATE_CONFIG: GateConfig;
+/** The delivery-gate configuration schema (the pluggable checklist + knobs). */
+export declare const GateConfigSchema: Schema<GateConfig>;
+/**
+ * Validate the gate configuration fail-loud: checklist ids unique and
+ * non-empty, valid spec-dimension bindings, bounded thresholds, compilable
+ * coverage regex, and sound reviewer knobs. A bad gate config must not
+ * silently produce an uncheckable checklist.
+ * @param gate - the gate config block.
+ */
+export declare function validateGateConfig(gate: GateConfig): void;
+/**
+ * Run the full delivery gate over the agent's session and settle the state.
+ * The deterministic phases fold first; the two reviewer phases run in
+ * parallel. Failures degrade to honest skip notices, never to throws.
+ * @param ctx - plugin context (subagents, commands, fs seams).
+ * @param config - the gate configuration.
+ * @param detection - the compiled test-run detection knobs.
+ * @param agent - the gated agent.
+ * @param signal - the caller's cancellation signal.
+ * @param language - prose language for the reviewer tasks.
+ * @returns the settled gate state.
+ */
+export declare function runGate(ctx: Context, config: GateConfig, detection: TestRunDetection, agent: Agent, signal: AbortSignal, language: ProseLanguage): Promise<GateState>;
+/**
+ * Settle one gate run's side effects: the durable `doublecheck/gate` session
+ * event (when the host stamps `ignorable`), the observability announcement,
+ * and the workspace `gate-report.md` copy. Failures of the file write never
+ * fail the run — the command result and the durable event remain.
+ * @param ctx - plugin context.
+ * @param config - the gate configuration.
+ * @param state - the settled gate state.
+ * @param agent - the gated agent.
+ * @param signal - the caller's cancellation signal.
+ * @param stampsIgnorable - whether the host stamps the `ignorable` marker.
+ */
+export declare function settleGate(ctx: Context, config: GateConfig, state: GateState, agent: Agent, signal: AbortSignal, stampsIgnorable: boolean): Promise<void>;
+/** The closures the `/gate` command reads; all live inside the guard's `apply` scope. */
+export interface GateCommandDeps {
+    config: GateConfig;
+    detection: TestRunDetection;
+    prose: GuardProse;
+    /** Runs the full gate and returns the settled state. */
+    runGate: (agent: Agent, signal: AbortSignal) => Promise<GateState>;
+    /** Settles the durable append, the announcement, and the workspace file. */
+    settleGate: (state: GateState, agent: Agent, signal: AbortSignal) => Promise<void>;
+    /** Whether the host stamps the `ignorable` marker (the durable write gate). */
+    stampsIgnorable: () => boolean;
+    /** The optional plan-mode service (weak seam: `ctx.get('planMode')`). */
+    planMode?: {
+        get(agent: Agent): {
+            active?: boolean;
+        } | undefined;
+    };
+}
+/**
+ * The `/gate` command handler: `status` renders the live checklist progress
+ * plus the latest settled run; `run` settles the full gate and returns the
+ * PR-ready markdown report; `config` renders the effective checklist.
+ * @param deps - the guard closures the handler reads.
+ * @returns the command handler for `ctx.commands.register`.
+ */
+export declare function gateHandler(deps: GateCommandDeps): (invocation: CommandInvocation) => CommandResult | Promise<CommandResult>;
+/**
+ * The injected gate-red notice message: short model-facing prose (role
+ * statement first) with the structured verdict and red count riding the
+ * durable `doublecheck-gate` message source, so the durable once-semantics
+ * and the report folds read the record without re-parsing the prose.
+ * @param verdict - the settled verdict.
+ * @param redCount - the red-item count.
+ * @param text - the model-facing notice text.
+ * @returns the injection-ready user message.
+ */
+export declare function gateInjection(verdict: GateVerdict, redCount: number, text: string): UserMessage;
+/**
+ * Render the effective gate configuration as markdown — the `/gate config`
+ * output: the pluggable checklist, the thresholds, and the reviewer knobs.
+ * Configuration contains no secrets by construction; nothing is redacted.
+ * @param config - the gate configuration.
+ * @returns the configuration document.
+ */
+export declare function renderGateConfigMarkdown(config: GateConfig): string;