diff --git a/CHANGELOG.md b/CHANGELOG.md index 679805e64..f8d22287c 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,5 +1,17 @@ # Changelog +## [1.91.124.0] - 2026-10-17 + +**/cso runs finish on time.** In the first paid runtime qualification, one comprehensive run submitted its findings ten minutes in, kept working for another 25 minutes without calling the helper, and was killed before it called `finish`. Its report was never finished, so the cold-start gate scored 63/64. The agent had no clock to follow. + +### Added +- **Helper replies carry `remainingSeconds`.** `start`, `resume`, `submit`, `scan`, `scanner-outcome`, `import-sarif`, `record-review`, `test-plan`, `runtime-plan`, `verify` and `inspect` report the whole seconds of investigation time left. The count stops where `scan`, `verify` and `record-review` start refusing with DEADLINE, so the final reporting minute stays free for `submit` and `finish`. The field is added after the command and only omitted if the deadline can't be read, so it never turns a saved change into an error. +- **The /cso skill follows the clock.** It now says to start only work that fits and, below 120 seconds, to stop waiting and `finish`. +- **Producer receipts record a content-free tool timeline.** For each tool call the eval producer records the name, its start and end offsets, and whether a subagent made it. It never records inputs or outputs. The timeline is capped at 4096 entries, validated on read, and left out of the batch index. Scoring is unchanged. + +### If you… +- **…drive `gstack-cso` from your own agent:** read `remainingSeconds` from each reply and stop starting work below two minutes. `inspect` refreshes it without side effects. + ## [1.91.123.0] - 2026-10-17 **/cso comprehensive repairs run again.** In the first paid runtime qualification, the agent recorded 148 repair reviews across 42 comprehensive runs, but not one verification attempt started. Every runtime scored 0/24 on held-out repairs. Two helper bugs stacked: the patch write failed on the snapshot's read-only files, and `runtime-plan` handed the agent a profile name that `verify` refused. diff --git a/VERSION b/VERSION index 48617770e..114fc58b8 100644 --- a/VERSION +++ b/VERSION @@ -1 +1 @@ -1.91.123.0 +1.91.124.0 diff --git a/agents-digest/gstack-AGENTS.md b/agents-digest/gstack-AGENTS.md index 957e4470a..9a47949cd 100644 --- a/agents-digest/gstack-AGENTS.md +++ b/agents-digest/gstack-AGENTS.md @@ -1,4 +1,4 @@ -# gstack digest v1.91.123.0 — regenerate/re-copy after upgrading gstack +# gstack digest v1.91.124.0 — regenerate/re-copy after upgrading gstack Behavioral rules from gstack (https://github.com/garrytan/gstack), compressed for agent hosts without a full skill install. The full skills add workflows, diff --git a/cso/SKILL.md b/cso/SKILL.md index 46e19240b..47293f45f 100644 --- a/cso/SKILL.md +++ b/cso/SKILL.md @@ -99,7 +99,7 @@ Start first; inspect snapshot, readiness, deadline, transformations, and coverag When updating coverage, copy every record's `domain` and `scope` exactly from `inspect`; a new scope leaves the planned scope unassessed. Only helper commands may update helper-owned records. -Budgets are ten minutes daily and thirty comprehensive, including one minute for reporting. Use at most three investigation workers and two reproduction groups per Docker endpoint. Attempts get five minutes and three harness repairs per finding, within the run deadline. Prioritize unresolved high impact. Report model usage only when exposed. +Budgets: ten minutes daily, thirty comprehensive. Use at most three investigation workers and two reproduction groups per Docker endpoint. Attempts get five minutes and three harness repairs per finding. Prioritize unresolved high impact. Report model usage only when exposed. Helper replies carry investigation `remainingSeconds`: start only what fits; below 120, stop and `finish`. --- ## Section index — Read each section when its situation applies diff --git a/cso/SKILL.md.tmpl b/cso/SKILL.md.tmpl index 9f936311a..765027bc5 100644 --- a/cso/SKILL.md.tmpl +++ b/cso/SKILL.md.tmpl @@ -99,7 +99,7 @@ Start first; inspect snapshot, readiness, deadline, transformations, and coverag When updating coverage, copy every record's `domain` and `scope` exactly from `inspect`; a new scope leaves the planned scope unassessed. Only helper commands may update helper-owned records. -Budgets are ten minutes daily and thirty comprehensive, including one minute for reporting. Use at most three investigation workers and two reproduction groups per Docker endpoint. Attempts get five minutes and three harness repairs per finding, within the run deadline. Prioritize unresolved high impact. Report model usage only when exposed. +Budgets: ten minutes daily, thirty comprehensive. Use at most three investigation workers and two reproduction groups per Docker endpoint. Attempts get five minutes and three harness repairs per finding. Prioritize unresolved high impact. Report model usage only when exposed. Helper replies carry investigation `remainingSeconds`: start only what fits; below 120, stop and `finish`. --- {{SECTION_INDEX:cso}} diff --git a/lib/cso/cli.ts b/lib/cso/cli.ts index 672272742..4cac4e3f8 100644 --- a/lib/cso/cli.ts +++ b/lib/cso/cli.ts @@ -57,6 +57,7 @@ import { readJson, reclaimDeadAttemptScratch, repoId, + executionDeadline, requireTime, retention, runDirectory, @@ -1609,6 +1610,7 @@ async function inspect(args: string[]) { ? readJson(join(dir, 'preparation.json')) : undefined, recovery: recoveryEvents(dir), + remainingSeconds: remainingSeconds(report.deadline), }); } async function read(args: string[]) { @@ -2796,12 +2798,31 @@ async function replay(args: string[], dependencies: CsoCliDependencies) { }); } +/** In-run replies that report the investigation time left; every one but `start` takes the run ID first. */ +const RUN_CLOCK_COMMANDS = new Set([ + 'start', + 'resume', + 'submit', + 'scan', + 'scanner-outcome', + 'import-sarif', + 'record-review', + 'test-plan', + 'runtime-plan', + 'verify', +]); +/** Whole seconds until the investigation deadline (the final reporting minute excluded), or undefined if unknown. */ +function remainingSeconds(deadline: string): number | undefined { + const ends = executionDeadline({ deadline }); + return Number.isFinite(ends) ? Math.max(0, Math.floor((ends - Date.now()) / 1000)) : undefined; +} export async function dispatchCsoCommand( command: string, args: string[], dependencies: CsoCliDependencies, ): Promise { args = [...args]; + const runId = args[0]; if (!['start', 'recheck', 'replay', 'doctor', 'provision-images'].includes(command)) { const started = Date.now(); retention(started, { deadlineMs: started + RETENTION_MAINTENANCE_MS, maxEntries: RETENTION_MAX_ENTRIES }); @@ -2876,7 +2897,18 @@ export async function dispatchCsoCommand( default: throw new CsoError('INVALID_ARGUMENT', `Unknown command: ${command}`); } - return result; + if (!RUN_CLOCK_COMMANDS.has(command) || !result || typeof result !== 'object' || Array.isArray(result)) + return result; + const deadline = (result as { deadline?: unknown }).deadline; + let seconds: number | undefined; + try { + seconds = remainingSeconds( + typeof deadline === 'string' ? deadline : loadReport(runDirectory(runId)).deadline, + ); + } catch { + seconds = undefined; + } + return seconds === undefined ? result : { ...result, remainingSeconds: seconds }; } const PRODUCTION_CLI_DEPENDENCIES: CsoCliDependencies = Object.freeze({ runtimeCatalog: RUNTIME_CATALOG, diff --git a/lib/cso/state.ts b/lib/cso/state.ts index 66ac7f6eb..aaaca9f79 100644 --- a/lib/cso/state.ts +++ b/lib/cso/state.ts @@ -944,7 +944,7 @@ export function loadReport(dir: string): RunReportV3 { export function event(report: RunReportV3, kind: string, message: string): void { report.events.push({ at: new Date().toISOString(), kind, message: redact(message) }); } -export function executionDeadline(report: RunReportV3): number { +export function executionDeadline(report: Pick): number { return Date.parse(report.deadline) - 60_000; } export function requireTime(report: RunReportV3): void { diff --git a/package.json b/package.json index a130d5382..1b48d7977 100644 --- a/package.json +++ b/package.json @@ -1,6 +1,6 @@ { "name": "gstack", - "version": "1.91.123", + "version": "1.91.124", "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.", "license": "MIT", "type": "module", diff --git a/scripts/cso-eval-producer.ts b/scripts/cso-eval-producer.ts index 333e71268..a2995473b 100644 --- a/scripts/cso-eval-producer.ts +++ b/scripts/cso-eval-producer.ts @@ -686,6 +686,7 @@ export async function runProducerCell(inputPath: string, receiptPath: string, op durationMs: run.durationMs, firstUsefulResultMs: null, toolCalls: run.toolCalls, + ...(run.toolTimeline ? { toolTimeline: run.toolTimeline } : {}), output, outputHash: sha256(output), usage: { diff --git a/scripts/cso-eval-protocol.ts b/scripts/cso-eval-protocol.ts index 88174ecff..58e76d50c 100644 --- a/scripts/cso-eval-protocol.ts +++ b/scripts/cso-eval-protocol.ts @@ -130,6 +130,8 @@ export interface ProducerReceipt { durationMs: number; firstUsefulResultMs: null; toolCalls: number; + /** Content-free tool timeline (name, offset, thread) for diagnosing overruns; absent when the adapter has none. */ + toolTimeline?: Array<{ atMs: number; endMs?: number; tool: string; agent: 'main' | 'subagent' }>; output: string; outputHash: string; usage: { @@ -148,7 +150,7 @@ export interface ProducerReceipt { * Compact trusted index; raw output remains in the separately retained receipt. * Collection adds the matrix cell that the receipt's cellRef names. */ -export type ProducerReceiptIndex = Omit & { +export type ProducerReceiptIndex = Omit & { cell: ProducerCell; error?: { code: string }; }; diff --git a/scripts/cso-eval.ts b/scripts/cso-eval.ts index ad02a9d2f..eef53751d 100644 --- a/scripts/cso-eval.ts +++ b/scripts/cso-eval.ts @@ -400,6 +400,8 @@ function validateReceipt(receipt: ProducerReceipt, cell: EvalCell, inputHash: st validateInstallationIdentity(receipt.installationIdentity, 'INVALID_PRODUCER_RECEIPT'); validateProviderIdentity(receipt.providerIdentity, cell.host === 'codex' ? 'gpt' : cell.host, 'INVALID_PRODUCER_RECEIPT'); validateArtifactInventory(receipt.artifacts, 'INVALID_PRODUCER_RECEIPT'); + const timeline = receipt.toolTimeline; + if (timeline !== undefined && (!Array.isArray(timeline) || timeline.length > 4096 || timeline.some(entry => !entry || !Number.isSafeInteger(entry.atMs) || entry.atMs < 0 || (entry.endMs !== undefined && (!Number.isSafeInteger(entry.endMs) || entry.endMs < entry.atMs)) || typeof entry.tool !== 'string' || !/^[A-Za-z][A-Za-z0-9_]{0,63}$/.test(entry.tool) || !['main', 'subagent'].includes(entry.agent) || Object.keys(entry).some(key => !['atMs', 'endMs', 'tool', 'agent'].includes(key))))) throw new Error('INVALID_PRODUCER_RECEIPT'); } function validateInstallationIdentity(identity: ProducerInstallationIdentity, error: string): void { @@ -530,7 +532,7 @@ export function collectProducerReceipts(matrix: EvalMatrix, schedule: PreparedEv estimatedCost: { measured: costs.length, denominator: expected.length, totalUSD: costs.length ? costs.reduce((sum, value) => sum + value, 0) : null, source: 'pricing-table-estimate' }, }); } - const indexes: ProducerReceiptIndex[] = bound.map(({ output: _output, error, ...receipt }) => ({ ...receipt, ...(error ? { error: { code: error.code } } : {}) })); + const indexes: ProducerReceiptIndex[] = bound.map(({ output: _output, toolTimeline: _toolTimeline, error, ...receipt }) => ({ ...receipt, ...(error ? { error: { code: error.code } } : {}) })); const base = { schemaVersion: 1 as const, matrixHash: matrixHash(matrix), scheduleHash: digest(JSON.stringify(schedule)), receipts: indexes.sort((left, right) => left.cell.id.localeCompare(right.cell.id)), summary: { scheduled: matrix.cells.length, prepared: schedule.preparedCells, submitted: bound.length, missing: matrix.cells.length - bound.length, matchedPairsExpected: expectedMatchedPairs(matrix), matchedPairsSubmitted: completePairs.length, modelMismatches, groups, diff --git a/test/claude-producer-stream.test.ts b/test/claude-producer-stream.test.ts index badfbe836..014801c12 100644 --- a/test/claude-producer-stream.test.ts +++ b/test/claude-producer-stream.test.ts @@ -33,7 +33,7 @@ describe('Claude producer stream accounting', () => { test('takes usage and turns from the final result and keeps the paid result contract', () => { const state = claudeStreamState(); for (const line of lines(start('a', 1000), delta(100), stop, result({}))) consumeClaudeStreamLine(state, line); - expect(resultFromClaudeStream(state, { model: 'claude-test', durationMs: 5 })).toEqual({ output: 'REPORT', tokens: { input: 2001, output: 201, cached: 14 }, durationMs: 5, toolCalls: 2, modelUsed: 'claude-test' }); + expect(resultFromClaudeStream(state, { model: 'claude-test', durationMs: 5 })).toEqual({ output: 'REPORT', tokens: { input: 2001, output: 201, cached: 14 }, durationMs: 5, toolCalls: 2, toolTimeline: [], modelUsed: 'claude-test' }); for (const ending of [[], [result({ result: ' ' })], [result({ result: 7 })], [result({ is_error: true })], [result({ subtype: 'error_max_turns' })]]) { const partial = claudeStreamState(); for (const line of lines(start('a', 10), delta(3), stop, ...ending)) consumeClaudeStreamLine(partial, line); @@ -52,7 +52,31 @@ describe('Claude producer stream accounting', () => { for (const line of lines(start('a', 1000), delta(100), stop, subagent('s1', 50000), subagent('s1', 50000), subagent('s2', 50001), result({ result: 'MAIN ENDED TURN', modelUsage: sessionUsage }), start('b', 1002), delta(102), stop, result({ num_turns: 1, result: 'REPORT', usage: { input_tokens: 1002, output_tokens: 102, cache_read_input_tokens: 7 }, modelUsage: sessionUsage }))) consumeClaudeStreamLine(state, line); - expect(resultFromClaudeStream(state, { model: 'claude-test', durationMs: 5 })).toEqual({ output: 'REPORT', tokens: { input: 153009, output: 15308, cached: 49 }, durationMs: 5, toolCalls: 3, modelUsed: 'claude-test' }); + expect(resultFromClaudeStream(state, { model: 'claude-test', durationMs: 5 })).toEqual({ output: 'REPORT', tokens: { input: 153009, output: 15308, cached: 49 }, durationMs: 5, toolCalls: 3, toolTimeline: [], modelUsed: 'claude-test' }); + }); + + test('records a content-free tool timeline: main and subagent calls once each, never their inputs', () => { + const toolStart = (id: string, name: string) => ({ type: 'stream_event', parent_tool_use_id: null, event: { type: 'content_block_start', index: 1, content_block: { type: 'tool_use', id, name, input: {} } } }); + const mainCall = { type: 'assistant', message: { id: 'a', content: [{ type: 'tool_use', id: 'toolu_bash', name: 'Bash', input: { command: 'gstack-cso inspect SECRET_PATH' } }] } }; + const subCall = { type: 'assistant', parent_tool_use_id: 'toolu_agent', message: { id: 's1', content: [{ type: 'tool_use', id: 'toolu_sub', name: 'Bash', input: { command: 'cat SECRET_PATH' } }], usage: { input_tokens: 1, output_tokens: 1 } } }; + const state = claudeStreamState(); + for (const line of lines(start('a', 10), toolStart('toolu_bash', 'Bash'), toolStart('toolu_agent', 'Agent'), mainCall, subCall, subCall, toolStart('toolu_odd', 'mcp__x/../y'))) consumeClaudeStreamLine(state, line); + const timeline = resultFromClaudeStream(state, { model: 'claude-test', durationMs: 5, error: { code: 'timeout', reason: 'exceeded 5ms' } }).toolTimeline!; + expect(timeline.map(({ tool, agent }) => [tool, agent])).toEqual([['Bash', 'main'], ['Agent', 'main'], ['Bash', 'subagent'], ['other', 'main']]); + expect(timeline.every(entry => Number.isSafeInteger(entry.atMs) && entry.atMs >= 0 && Object.keys(entry).length === 3)).toBe(true); + expect(JSON.stringify(timeline)).not.toContain('SECRET_PATH'); + }); + + test('stamps a tool call end once from its tool_result and keeps the timeline and its id index at the cap', () => { + const state = claudeStreamState(); + const toolResult = (id: string) => ({ type: 'user', message: { content: [{ type: 'tool_result', tool_use_id: id, content: 'SECRET_OUTPUT' }] } }); + const call = (id: string) => ({ type: 'assistant', message: { id: `m${id}`, content: [{ type: 'tool_use', id, name: 'Bash', input: {} }] } }); + for (const line of lines(call('t0'), toolResult('t0'), toolResult('t0'), toolResult('unknown'), call('t1'))) consumeClaudeStreamLine(state, line); + expect(state.toolTimeline.map(entry => entry.endMs === undefined)).toEqual([false, true]); + expect(JSON.stringify(state.toolTimeline)).not.toContain('SECRET_OUTPUT'); + for (let i = 2; i < 5000; i++) consumeClaudeStreamLine(state, JSON.stringify(call(`t${i}`))); + expect(state.toolTimeline).toHaveLength(4096); + expect(state.toolUseIndex.size).toBe(4096); }); test('a killed run counts each subagent message once and never reports subagent text as the producer output', () => { diff --git a/test/cso-cli.test.ts b/test/cso-cli.test.ts index 4a14016df..b3b10735b 100644 --- a/test/cso-cli.test.ts +++ b/test/cso-cli.test.ts @@ -69,6 +69,17 @@ describe('compiled CSO command workflow',()=>{ runGit('init','-q');runGit('config','user.email','fixture@example.test');runGit('config','user.name','Fixture');fs.writeFileSync(path.join(planRepo,'package.json'),JSON.stringify({name:'sensitive-plan',version:'1.0.0',scripts:{start:'node app.js',test:'node --test'}}));fs.writeFileSync(path.join(planRepo,'package-lock.json'),JSON.stringify({name:'sensitive-plan',version:'1.0.0',lockfileVersion:3,packages:{'':{name:'sensitive-plan',version:'1.0.0'}}}));fs.writeFileSync(path.join(planRepo,'app.js'),'export const app = true\n');fs.writeFileSync(path.join(planRepo,testPath),'const test=require("node:test");test("control",()=>{})\n');runGit('add','.');runGit('commit','-qm','sensitive test path'); const started=command(['start','--repo',planRepo,'--comprehensive','--offline']);expect(started.status).toBe(0);const run=JSON.parse(started.stdout),planned=command(['runtime-plan',run.runId,'node','--port','3456']);expect(planned.status).toBe(0);const plan=JSON.parse(planned.stdout),serialized=JSON.stringify(plan); expect(serialized).not.toContain(wallet);const testHandle=plan.tests.files.find((value:string)=>value.startsWith('@cso-path//'));expect(testHandle).toMatch(/^@cso-path\/\/[a-f0-9]{32}$/);expect(plan.tests.commands[0].args).toContain(`./${testHandle}`);expect(plan.start.command).toEqual({executable:'/usr/local/bin/node',args:['app.js']}); + const budget=(Date.parse(run.deadline)-Date.parse(run.createdAt))/1000;for(const remaining of [run.remainingSeconds,plan.remainingSeconds]){expect(Number.isSafeInteger(remaining)).toBe(true);expect(remaining).toBeGreaterThan(budget-120);expect(remaining).toBeLessThanOrEqual(budget-60);}expect(plan.remainingSeconds).toBeLessThanOrEqual(run.remainingSeconds); + }); + test('run clock counts to the investigation deadline: zero in the reporting minute, which still accepts submit until the audit deadline',()=>{ + const started=command(['start','--repo',repo,'--scope','auth','--offline']);expect(started.status).toBe(0);const run=JSON.parse(started.stdout),reportPath=path.join(state,'security','cso',run.repoId,run.runId,'report.json'),input=path.join(root,`clock-${run.runId}.json`);fs.writeFileSync(input,'{}'); + const submitted=command(['submit',run.runId,input]);expect(submitted.status).toBe(0);const inspected=command(['inspect',run.runId]);expect(inspected.status).toBe(0); + for(const remaining of [JSON.parse(submitted.stdout).remainingSeconds,JSON.parse(inspected.stdout).remainingSeconds]){expect(Number.isSafeInteger(remaining)).toBe(true);expect(remaining).toBeGreaterThan(0);expect(remaining).toBeLessThanOrEqual(run.remainingSeconds);} + const report=JSON.parse(fs.readFileSync(reportPath,'utf8'));report.deadline=new Date(Date.now()+30_000).toISOString();fs.writeFileSync(reportPath,JSON.stringify(report,null,2)+'\n'); + const late=command(['inspect',run.runId]);expect(late.status).toBe(0);expect(JSON.parse(late.stdout).remainingSeconds).toBe(0); + const reporting=command(['submit',run.runId,input]);expect(reporting.status).toBe(0);expect(JSON.parse(reporting.stdout).remainingSeconds).toBe(0); + report.deadline=new Date(Date.now()-1_000).toISOString();fs.writeFileSync(reportPath,JSON.stringify({...JSON.parse(fs.readFileSync(reportPath,'utf8')),deadline:report.deadline},null,2)+'\n'); + const refused=command(['submit',run.runId,input]);expect(refused.status).not.toBe(0);expect(refused.stderr).toContain('DEADLINE');expect(refused.stdout).toBe(''); }); test('wallet-shaped helper finding IDs survive submit, review, and verification admission',()=>{const started=command(['start','--repo',repo,'--comprehensive','--offline']);expect(started.status).toBe(0);const run=JSON.parse(started.stdout),base:any={title:'Wallet-shaped helper ID',rootCause:'Tenant query omits caller tenant predicate 0',location:{path:'src/users.ts',line:1,symbol:'tenantQuery'},advisoryIds:[],severity:'high',confidence:'high',confidenceRationale:'Caller-to-sink trace directly establishes the missing predicate',evidence:'supported',attackerControl:'Tenant chooses the record ID',impact:'Cross-tenant record disclosure',scenario:'A tenant requests another tenant record by ID',trace:['route','tenantQuery','database'],references:['src/users.ts:1'],recommendation:'Bind the query to the authenticated tenant',challenge:{reviewer:'reviewer-session',independent:true,mode:'independent_agent',callers:'route caller checked',controls:'authentication lacks authorization',counterevidence:'none found',conclusion:'tenant predicate is absent'}};let helperId='';for(let i=0;i<10_000;i++){base.rootCause=`Tenant query omits caller tenant predicate ${i}`;helperId=fingerprint(base);if(redact(helperId)!==helperId)break;}expect(redact(helperId)).not.toBe(helperId);const input=path.join(root,`wallet-id-${run.runId}.json`);fs.writeFileSync(input,JSON.stringify({findings:[base]}));expect(command(['submit',run.runId,input]).status).toBe(0);const reportPath=path.join(state,'security','cso',run.repoId,run.runId,'report.json'),stored=JSON.parse(fs.readFileSync(reportPath,'utf8'));expect(stored.findings[0].id).toBe(helperId);const before=fs.readFileSync(path.join(repo,'src','users.ts'),'utf8'),request:any={findingId:helperId,runtimeProfile:'node-v1',port:3456,start:{executable:'/usr/local/bin/node',args:['src/users.ts']},legitimate:[{name:'control',path:'/health',method:'GET',expected:{status:200}}],security:{name:'tenant boundary',path:'/users/other',method:'GET',expected:{status:403},vulnerable:{status:200}},existingTests:[{executable:'/usr/local/bin/node',args:['--test']}],testFiles:['src/control.test.ts'],fixtures:{},boundaryFiles:['src/users.ts'],changes:[{path:'src/users.ts',beforeSha256:sha256(before),after:before+'// tenant predicate\n',effect:'source'}],review:{reviewer:'reviewer-session',independent:true,rootCauseRepaired:true,featurePreserved:true,boundaryMocks:false,rationale:'The tenant predicate is added while the control remains intact',reviewedPatchHash:''}};request.review.reviewedPatchHash=patchHash(request);fs.writeFileSync(input,JSON.stringify(request));const reviewed=command(['record-review',run.runId,input,'--producer','producer-session']);expect(reviewed.status).toBe(0);request.review.artifactId=JSON.parse(reviewed.stdout).reviewArtifactId;fs.writeFileSync(input,JSON.stringify(request));const verified=command(['verify',run.runId,input]);expect(verified.status).not.toBe(0);expect(verified.stderr).toContain('PREREQUISITE');expect(verified.stderr).not.toContain('REDACTION_FAILED');const review=JSON.parse(fs.readFileSync(path.join(path.dirname(reportPath),'reviews',`${request.review.artifactId}.json`),'utf8'));expect(review.findingId).toBe(helperId);}); test('daily find, early submit, and finish persist a truthful report without changing source',()=>{const before=git('status','--porcelain=v1','-z'),started=command(['start','--repo',repo,'--scope','auth','--offline']);expect(started.status).toBe(0);const run=JSON.parse(started.stdout),runId=run.runId;expect(run.completeness).toBe('not assessed');const reportPath=path.join(state,'security','cso',run.repoId,runId,'report.json');expect(fs.existsSync(reportPath)).toBe(true);const current=JSON.parse(fs.readFileSync(reportPath,'utf8')); diff --git a/test/cso-eval.test.ts b/test/cso-eval.test.ts index 6d6559576..34b773c84 100644 --- a/test/cso-eval.test.ts +++ b/test/cso-eval.test.ts @@ -268,6 +268,11 @@ describe('CSO matched evaluation accounting', () => { artifactTamper.artifacts.totalBytes = 1; expect(() => collectProducerReceipts(matrix, syntheticSchedule(), [rehashReceipt(artifactTamper)])).toThrow('INVALID_PRODUCER_RECEIPT'); + const timed = rehashReceipt({ ...structuredClone(first), toolTimeline: [{ atMs: 5, endMs: 9, tool: 'Bash', agent: 'main' }, { atMs: 7, tool: 'Agent', agent: 'subagent' }] }); + expect(collectProducerReceipts(matrix, syntheticSchedule(), [timed]).receipts[0]).not.toHaveProperty('toolTimeline'); + for (const entry of [{ atMs: 9, endMs: 5, tool: 'Bash', agent: 'main' }, { atMs: 1, tool: 'cat /secret', agent: 'main' }, { atMs: 1, tool: 'Bash', agent: 'main', input: 'x' }]) + expect(() => collectProducerReceipts(matrix, syntheticSchedule(), [rehashReceipt({ ...structuredClone(first), toolTimeline: [entry] } as ProducerReceipt)])).toThrow('INVALID_PRODUCER_RECEIPT'); + const mixedInstallation = syntheticReceipt(cells[1]); mixedInstallation.installationIdentity = syntheticInstallationIdentity('different-installation'); expect(() => collectProducerReceipts(matrix, syntheticSchedule(), [first, rehashReceipt(mixedInstallation)])).toThrow('UNMATCHED_PRODUCER_INSTALLATIONS'); diff --git a/test/helpers/providers/claude-stream.ts b/test/helpers/providers/claude-stream.ts index 18374b592..5cdb1da6e 100644 --- a/test/helpers/providers/claude-stream.ts +++ b/test/helpers/providers/claude-stream.ts @@ -1,5 +1,5 @@ import { spawn } from 'child_process'; -import type { RunResult } from './types'; +import type { RunResult, ToolTimelineEntry } from './types'; /** Usage, turns and the latest text from `claude -p --output-format stream-json --verbose --include-partial-messages`. */ export interface ClaudeStreamState { @@ -14,14 +14,36 @@ export interface ClaudeStreamState { lastText: string; result?: Record; invalidLines: number; + startedAt: number; + /** Content-free: each tool call's name, start and end offsets and thread, never its input or output. */ + toolTimeline: ToolTimelineEntry[]; + /** Timeline index by tool-use id, for recorded calls only, so it stays within the timeline cap. */ + toolUseIndex: Map; } export function claudeStreamState(): ClaudeStreamState { - return { totals: { input: 0, output: 0, cached: 0 }, open: new Map(), subagents: new Map(), turns: 0, resultTurns: 0, lastText: '', invalidLines: 0 }; + return { totals: { input: 0, output: 0, cached: 0 }, open: new Map(), subagents: new Map(), turns: 0, resultTurns: 0, lastText: '', invalidLines: 0, startedAt: Date.now(), toolTimeline: [], toolUseIndex: new Map() }; } +const TOOL_TIMELINE_LIMIT = 4096; + const count = (value: unknown): number => typeof value === 'number' && Number.isSafeInteger(value) && value >= 0 ? value : 0; +function recordToolUse(state: ClaudeStreamState, block: { id?: unknown; name?: unknown }, agent: 'main' | 'subagent'): void { + if (typeof block.id !== 'string' || state.toolUseIndex.has(block.id) || state.toolTimeline.length >= TOOL_TIMELINE_LIMIT) return; + state.toolUseIndex.set(block.id, state.toolTimeline.length); + state.toolTimeline.push({ + atMs: Date.now() - state.startedAt, + tool: typeof block.name === 'string' && /^[A-Za-z][A-Za-z0-9_]{0,63}$/.test(block.name) ? block.name : 'other', + agent, + }); +} + +function recordToolResult(state: ClaudeStreamState, block: { tool_use_id?: unknown }): void { + const index = typeof block.tool_use_id === 'string' ? state.toolUseIndex.get(block.tool_use_id) : undefined; + if (index !== undefined && state.toolTimeline[index].endMs === undefined) state.toolTimeline[index].endMs = Date.now() - state.startedAt; +} + function close(state: ClaudeStreamState, key: string): void { const message = state.open.get(key); if (!message) return; @@ -39,6 +61,13 @@ export function consumeClaudeStreamLine(state: ClaudeStreamState, line: string): if (Number.isSafeInteger(event.num_turns) && event.num_turns >= 0) state.resultTurns += event.num_turns; return; } + if (event.type === 'assistant' && Array.isArray(event.message?.content)) { + for (const block of event.message.content) if (block?.type === 'tool_use') recordToolUse(state, block, typeof event.parent_tool_use_id === 'string' ? 'subagent' : 'main'); + } + if (event.type === 'user' && Array.isArray(event.message?.content)) { + for (const block of event.message.content) if (block?.type === 'tool_result') recordToolResult(state, block); + return; + } if (event.type === 'assistant' && typeof event.parent_tool_use_id === 'string') { const usage = event.message?.usage; if (typeof event.message?.id === 'string' && usage) state.subagents.set(event.message.id, { input: count(usage.input_tokens), output: count(usage.output_tokens), cached: count(usage.cache_read_input_tokens) }); @@ -56,6 +85,8 @@ export function consumeClaudeStreamLine(state: ClaudeStreamState, line: string): const usage = stream.message?.usage ?? {}; state.open.set(key, { input: count(usage.input_tokens), output: count(usage.output_tokens), cached: count(usage.cache_read_input_tokens) }); if (key === 'main') state.turns++; + } else if (stream.type === 'content_block_start' && stream.content_block?.type === 'tool_use') { + recordToolUse(state, stream.content_block, key === 'main' ? 'main' : 'subagent'); } else if (stream.type === 'message_delta' && stream.usage) { const message = state.open.get(key); if (!message) return; @@ -99,7 +130,7 @@ export function resultFromClaudeStream(state: ClaudeStreamState, opts: { model: const tokens = (result && resultTokens(result)) ?? claudeStreamUsage(state); const toolCalls = result ? state.resultTurns : state.turns; const modelUsed = typeof result?.model === 'string' && result.model ? result.model : opts.model; - const base = { tokens, durationMs: opts.durationMs, toolCalls, modelUsed }; + const base = { tokens, durationMs: opts.durationMs, toolCalls, toolTimeline: state.toolTimeline, modelUsed }; if (opts.error) return { ...base, output: typeof result?.result === 'string' ? result.result : state.lastText, error: opts.error }; const explicitError = result?.is_error === true || (typeof result?.subtype === 'string' && result.subtype !== 'success'); if (!result || explicitError || typeof result.result !== 'string' || !result.result.trim()) { diff --git a/test/helpers/providers/types.ts b/test/helpers/providers/types.ts index 59063e7b2..e9b51527e 100644 --- a/test/helpers/providers/types.ts +++ b/test/helpers/providers/types.ts @@ -49,6 +49,14 @@ export type RunError = | 'binary_missing' // CLI not found on PATH. | 'unknown'; // Catch-all with reason populated. +export interface ToolTimelineEntry { + atMs: number; + /** Set when the call's tool_result arrived; absent for a call still running when the stream ended. */ + endMs?: number; + tool: string; + agent: 'main' | 'subagent'; +} + export interface RunResult { /** Provider's textual output for the prompt. */ output: string; @@ -58,6 +66,8 @@ export interface RunResult { durationMs: number; /** Count of tool/function calls made during the run (0 if unsupported). */ toolCalls: number; + /** Tool calls in start order: name, start and end offsets from run start, and thread; no inputs or outputs. Unset if unsupported. */ + toolTimeline?: ToolTimelineEntry[]; /** Actual model ID the provider reports using (may be a variant of the family). */ modelUsed: string; /** If the run failed, error code + human reason. output/tokens may be partial. */