agent-afk 5.296.0 → 5.296.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/cli.mjs +341 -341
- package/dist/improve/eval-run/contracts.d.ts +5 -22
- package/dist/improve/eval-run/contracts.tfd.d.ts +4 -0
- package/dist/improve/eval-run/contracts.types.d.ts +20 -0
- package/dist/improve/eval-run/runner.d.ts +2 -12
- package/dist/improve/eval-run/runner.render.d.ts +13 -0
- package/dist/improve/propose/template-engine.d.ts +2 -0
- package/dist/improve/propose/template-engine.templates-density.d.ts +14 -0
- package/dist/improve/propose/template-engine.templates-loop.d.ts +14 -0
- package/dist/index.mjs +141 -141
- package/dist/skills/audit-fit/handler.d.ts +4 -0
- package/dist/skills/audit-fit/index.d.ts +3 -98
- package/dist/skills/audit-fit/schemas.d.ts +98 -0
- package/dist/telegram.mjs +164 -164
- package/package.json +1 -1
|
@@ -1,25 +1,8 @@
|
|
|
1
|
-
import {
|
|
2
|
-
import
|
|
3
|
-
export
|
|
4
|
-
|
|
5
|
-
|
|
6
|
-
}
|
|
7
|
-
export interface EvalContract {
|
|
8
|
-
id: string;
|
|
9
|
-
patternId: FailurePattern;
|
|
10
|
-
title: string;
|
|
11
|
-
run: () => Promise<ContractProbeResult>;
|
|
12
|
-
}
|
|
13
|
-
export declare function snapshot(value: unknown): string;
|
|
14
|
-
export declare function makeCheck(args: {
|
|
15
|
-
name: string;
|
|
16
|
-
description: string;
|
|
17
|
-
pass: boolean;
|
|
18
|
-
expected: string;
|
|
19
|
-
actual: string;
|
|
20
|
-
status?: EvalCheckStatus;
|
|
21
|
-
}): EvalCheck;
|
|
22
|
-
export declare function buildToolFailureClassificationCorpus(): SessionRead[];
|
|
1
|
+
import type { FailurePattern } from '../schemas.js';
|
|
2
|
+
import { type EvalContract } from './contracts.types.js';
|
|
3
|
+
export type { ContractProbeResult, EvalContract } from './contracts.types.js';
|
|
4
|
+
export { makeCheck, snapshot } from './contracts.types.js';
|
|
5
|
+
export { buildToolFailureClassificationCorpus } from './contracts.tfd.js';
|
|
23
6
|
export declare function resolveContract(patternId: FailurePattern): EvalContract | undefined;
|
|
24
7
|
export declare function supportedContractPatterns(): readonly FailurePattern[];
|
|
25
8
|
export declare function knownContractIds(): readonly string[];
|
|
@@ -0,0 +1,4 @@
|
|
|
1
|
+
import { type SessionRead } from '../scan/reader.js';
|
|
2
|
+
import { type ContractProbeResult } from './contracts.types.js';
|
|
3
|
+
export declare function buildToolFailureClassificationCorpus(): SessionRead[];
|
|
4
|
+
export declare function runToolFailureDensityContract(): Promise<ContractProbeResult>;
|
|
@@ -0,0 +1,20 @@
|
|
|
1
|
+
import type { EvalCheck, EvalCheckStatus, EvalRunEvidenceRef, FailurePattern } from '../schemas.js';
|
|
2
|
+
export interface ContractProbeResult {
|
|
3
|
+
checks: EvalCheck[];
|
|
4
|
+
evidence: EvalRunEvidenceRef[];
|
|
5
|
+
}
|
|
6
|
+
export interface EvalContract {
|
|
7
|
+
id: string;
|
|
8
|
+
patternId: FailurePattern;
|
|
9
|
+
title: string;
|
|
10
|
+
run: () => Promise<ContractProbeResult>;
|
|
11
|
+
}
|
|
12
|
+
export declare function snapshot(value: unknown): string;
|
|
13
|
+
export declare function makeCheck(args: {
|
|
14
|
+
name: string;
|
|
15
|
+
description: string;
|
|
16
|
+
pass: boolean;
|
|
17
|
+
expected: string;
|
|
18
|
+
actual: string;
|
|
19
|
+
status?: EvalCheckStatus;
|
|
20
|
+
}): EvalCheck;
|
|
@@ -1,6 +1,8 @@
|
|
|
1
1
|
import { type EvalCase, type EvalCheck, type EvalRun, type EvalRunStatus } from '../schemas.js';
|
|
2
2
|
import type { IdContext } from '../eval-gen/writer.js';
|
|
3
3
|
import { type LoopDriver } from './replay.js';
|
|
4
|
+
export type { EvalRunListEntry } from './runner.render.js';
|
|
5
|
+
export { listEvalRuns, getEvalRun, renderEvalRunMarkdown } from './runner.render.js';
|
|
4
6
|
export declare const EVAL_RUN_RUNNER_VERSION = "eval-run@v1";
|
|
5
7
|
export declare function generateEvalRunId(cardSlug: string, ctx?: IdContext): string;
|
|
6
8
|
export interface RunEvalCaseContext {
|
|
@@ -23,15 +25,3 @@ export interface WriteEvalRunOutcome {
|
|
|
23
25
|
markdownPath: string;
|
|
24
26
|
}
|
|
25
27
|
export declare function writeEvalRun(evalRun: EvalRun): WriteEvalRunOutcome;
|
|
26
|
-
export declare function renderEvalRunMarkdown(run: EvalRun): string;
|
|
27
|
-
export interface EvalRunListEntry {
|
|
28
|
-
evalRunId: string;
|
|
29
|
-
evalCaseId: string;
|
|
30
|
-
cardSlug: string;
|
|
31
|
-
patternId: EvalRun['patternId'];
|
|
32
|
-
contract: string | null;
|
|
33
|
-
status: EvalRunStatus;
|
|
34
|
-
createdAt: string;
|
|
35
|
-
}
|
|
36
|
-
export declare function listEvalRuns(): EvalRunListEntry[];
|
|
37
|
-
export declare function getEvalRun(evalRunId: string): EvalRun | undefined;
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
import { type EvalRun, type EvalRunStatus } from '../schemas.js';
|
|
2
|
+
export declare function renderEvalRunMarkdown(run: EvalRun): string;
|
|
3
|
+
export interface EvalRunListEntry {
|
|
4
|
+
evalRunId: string;
|
|
5
|
+
evalCaseId: string;
|
|
6
|
+
cardSlug: string;
|
|
7
|
+
patternId: EvalRun['patternId'];
|
|
8
|
+
contract: string | null;
|
|
9
|
+
status: EvalRunStatus;
|
|
10
|
+
createdAt: string;
|
|
11
|
+
}
|
|
12
|
+
export declare function listEvalRuns(): EvalRunListEntry[];
|
|
13
|
+
export declare function getEvalRun(evalRunId: string): EvalRun | undefined;
|
|
@@ -1,7 +1,9 @@
|
|
|
1
1
|
import { type FailureCard, type ImprovementProposal, type LikelyFile, type Severity } from '../schemas.js';
|
|
2
|
+
import { closureAdviceFor as _closureAdviceFor } from './template-engine.templates-density.js';
|
|
2
3
|
export interface TemplateContext {
|
|
3
4
|
proposalId: string;
|
|
4
5
|
now?: () => Date;
|
|
5
6
|
}
|
|
6
7
|
export declare function proposeFromCard(card: FailureCard, ctx: TemplateContext): ImprovementProposal;
|
|
7
8
|
export declare function deriveRiskLevel(floor: Severity, files: readonly LikelyFile[]): Severity;
|
|
9
|
+
export { _closureAdviceFor as closureAdviceFor };
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
import type { FailureCard, LikelyFile, ValidationPlan } from '../schemas.js';
|
|
2
|
+
import type { RootCauseClass, Severity } from '../schemas.js';
|
|
3
|
+
interface PatternTemplate {
|
|
4
|
+
rootCauseClass: RootCauseClass;
|
|
5
|
+
hypothesis(card: FailureCard): string;
|
|
6
|
+
fixSketch(card: FailureCard): string;
|
|
7
|
+
likelyFiles: readonly LikelyFile[];
|
|
8
|
+
riskFloor: Severity;
|
|
9
|
+
validationPlan: ValidationPlan;
|
|
10
|
+
}
|
|
11
|
+
export declare const toolFailureDensityTemplate: PatternTemplate;
|
|
12
|
+
export declare function closureAdviceFor(reason: string): string;
|
|
13
|
+
export declare const closureAnomalyTemplate: PatternTemplate;
|
|
14
|
+
export {};
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
import type { FailureCard, LikelyFile, ValidationPlan } from '../schemas.js';
|
|
2
|
+
import type { RootCauseClass, Severity } from '../schemas.js';
|
|
3
|
+
interface PatternTemplate {
|
|
4
|
+
rootCauseClass: RootCauseClass;
|
|
5
|
+
hypothesis(card: FailureCard): string;
|
|
6
|
+
fixSketch(card: FailureCard): string;
|
|
7
|
+
likelyFiles: readonly LikelyFile[];
|
|
8
|
+
riskFloor: Severity;
|
|
9
|
+
validationPlan: ValidationPlan;
|
|
10
|
+
}
|
|
11
|
+
export declare const repeatedToolUseTemplate: PatternTemplate;
|
|
12
|
+
export declare const subagentBlockTemplate: PatternTemplate;
|
|
13
|
+
export declare const subagentReadDenialTemplate: PatternTemplate;
|
|
14
|
+
export {};
|