@mjasnikovs/pi-task 0.18.4 → 0.18.6
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/config/config.d.ts +18 -0
- package/dist/config/config.js +10 -1
- package/dist/config/register.js +32 -5
- package/dist/task/accept-debt.d.ts +52 -0
- package/dist/task/accept-debt.js +0 -0
- package/dist/task/auto-orchestrator.d.ts +2 -0
- package/dist/task/auto-orchestrator.js +20 -0
- package/dist/task/final-gate.d.ts +8 -0
- package/dist/task/final-gate.js +27 -7
- package/dist/task/frozen-path-guard.d.ts +39 -0
- package/dist/task/frozen-path-guard.js +116 -0
- package/dist/task/gate-deps.js +25 -0
- package/dist/task/phases.d.ts +6 -2
- package/dist/task/phases.js +7 -2
- package/dist/task/repo-health-check.d.ts +11 -0
- package/dist/task/repo-health-check.js +26 -3
- package/dist/task/service-blocks.d.ts +2 -2
- package/dist/task/service-blocks.js +3 -1
- package/dist/task/task-gates.d.ts +24 -0
- package/dist/task/task-gates.js +78 -8
- package/dist/workers/brave-search.d.ts +2 -5
- package/dist/workers/brave-warning.d.ts +4 -6
- package/dist/workers/brave-warning.js +11 -10
- package/dist/workers/ddg-search.d.ts +24 -0
- package/dist/workers/ddg-search.js +130 -0
- package/dist/workers/exa-search.d.ts +24 -0
- package/dist/workers/exa-search.js +164 -0
- package/dist/workers/pi-worker-docs.js +13 -1
- package/dist/workers/pi-worker-fetch.js +10 -1
- package/dist/workers/pi-worker-search.d.ts +7 -1
- package/dist/workers/pi-worker-search.js +18 -5
- package/dist/workers/research-cache.d.ts +39 -0
- package/dist/workers/research-cache.js +140 -0
- package/dist/workers/search-core.d.ts +14 -2
- package/dist/workers/search-core.js +34 -2
- package/dist/workers/search-types.d.ts +15 -0
- package/dist/workers/search-types.js +4 -0
- package/dist/workers/shared.d.ts +17 -0
- package/dist/workers/shared.js +0 -0
- package/package.json +1 -1
package/dist/config/config.d.ts
CHANGED
|
@@ -1,3 +1,4 @@
|
|
|
1
|
+
import { type SearchProvider } from '../workers/search-types.js';
|
|
1
2
|
export interface PiTaskConfig {
|
|
2
3
|
remote: boolean;
|
|
3
4
|
compressReasoning: boolean;
|
|
@@ -12,6 +13,23 @@ export interface PiTaskConfig {
|
|
|
12
13
|
* phases.ts). Turn on only for a parallel-capable backend.
|
|
13
14
|
*/
|
|
14
15
|
parallelResearchWorkers: boolean;
|
|
16
|
+
/**
|
|
17
|
+
* Cache docs/search/fetch worker RESULTS for the duration of one /task-auto run
|
|
18
|
+
* so sibling tasks re-asking the same (package/url, query) reuse the first
|
|
19
|
+
* pipeline's digest instead of re-fetching (mx5 run-8 F10: the research phase
|
|
20
|
+
* burned 75 of 363 min largely re-fetching the same external docs across ~20
|
|
21
|
+
* siblings). Per-run isolated, external-only (project-source `.` lookups excluded),
|
|
22
|
+
* success-only. DEFAULT ON — the F10 live A/B showed no answer-quality regression
|
|
23
|
+
* (a cache hit serves byte-identical text to the first fetch; distinct queries never
|
|
24
|
+
* collide).
|
|
25
|
+
*/
|
|
26
|
+
researchCache: boolean;
|
|
27
|
+
/**
|
|
28
|
+
* Which engine backs web search (pi-worker-search + freshness/enrichment).
|
|
29
|
+
* `exa` and `ddg` need no API key; `brave` needs BRAVE_SEARCH_API_KEY.
|
|
30
|
+
* DEFAULT `exa` so search works out of the box with zero configuration.
|
|
31
|
+
*/
|
|
32
|
+
searchProvider: SearchProvider;
|
|
15
33
|
}
|
|
16
34
|
export declare function getConfig(): PiTaskConfig;
|
|
17
35
|
export declare function saveConfig(config: PiTaskConfig): Promise<void>;
|
package/dist/config/config.js
CHANGED
|
@@ -2,6 +2,7 @@ import * as fs from 'node:fs';
|
|
|
2
2
|
import * as fsp from 'node:fs/promises';
|
|
3
3
|
import * as path from 'node:path';
|
|
4
4
|
import * as os from 'node:os';
|
|
5
|
+
import { isSearchProvider } from '../workers/search-types.js';
|
|
5
6
|
const DEFAULTS = {
|
|
6
7
|
remote: true,
|
|
7
8
|
compressReasoning: true,
|
|
@@ -9,7 +10,11 @@ const DEFAULTS = {
|
|
|
9
10
|
orientation: true,
|
|
10
11
|
enforceGuidelines: true,
|
|
11
12
|
verifyWork: true,
|
|
12
|
-
parallelResearchWorkers: false
|
|
13
|
+
parallelResearchWorkers: false,
|
|
14
|
+
// ON: the F10 live A/B showed no answer-quality regression (fidelity 3/3, quality
|
|
15
|
+
// 3/3, 0 collisions; ~14.5s of repeated docs lookups collapse to 0ms on a hit).
|
|
16
|
+
researchCache: true,
|
|
17
|
+
searchProvider: 'exa'
|
|
13
18
|
};
|
|
14
19
|
const CONFIG_PATH = path.join(os.homedir(), '.config', 'pi-task', 'config.json');
|
|
15
20
|
const _g = globalThis;
|
|
@@ -23,6 +28,10 @@ if (!G.loaded) {
|
|
|
23
28
|
try {
|
|
24
29
|
const raw = fs.readFileSync(CONFIG_PATH, 'utf8');
|
|
25
30
|
const parsed = JSON.parse(raw);
|
|
31
|
+
// A hand-edited or stale enum value must not leak an unknown provider
|
|
32
|
+
// into the dispatch switch — fall back to the default.
|
|
33
|
+
if (!isSearchProvider(parsed.searchProvider))
|
|
34
|
+
delete parsed.searchProvider;
|
|
26
35
|
G.config = { ...DEFAULTS, ...parsed };
|
|
27
36
|
}
|
|
28
37
|
catch {
|
package/dist/config/register.js
CHANGED
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import { SettingsList, visibleWidth } from '@earendil-works/pi-tui';
|
|
2
2
|
import { registerBridgeCommand } from '../remote/bridge.js';
|
|
3
|
+
import { SEARCH_PROVIDERS, isSearchProvider } from '../workers/search-types.js';
|
|
3
4
|
import { getConfig, saveConfig } from './config.js';
|
|
4
5
|
const CONFIG_TITLE = 'pi-task settings';
|
|
5
6
|
/**
|
|
@@ -46,6 +47,10 @@ class BorderedBox {
|
|
|
46
47
|
this.child.handleInput(data);
|
|
47
48
|
}
|
|
48
49
|
}
|
|
50
|
+
/**
|
|
51
|
+
* Every setting rendered by /task-config. Boolean settings omit `values` and
|
|
52
|
+
* toggle on/off; enum settings list their values and cycle through them.
|
|
53
|
+
*/
|
|
49
54
|
const ITEMS = [
|
|
50
55
|
{ id: 'remote', label: 'remote', description: 'Remote UI server (QR code, phone access)' },
|
|
51
56
|
{
|
|
@@ -77,6 +82,17 @@ const ITEMS = [
|
|
|
77
82
|
id: 'parallelResearchWorkers',
|
|
78
83
|
label: 'parallel research',
|
|
79
84
|
description: 'Run the 4 research workers concurrently. Leave OFF on a single-GPU local server (serial is measurably faster there); turn on only for a parallel-capable model backend'
|
|
85
|
+
},
|
|
86
|
+
{
|
|
87
|
+
id: 'researchCache',
|
|
88
|
+
label: 'research cache',
|
|
89
|
+
description: 'Cache docs/search/fetch results within one /task-auto run so sibling tasks reuse the first pipeline’s digest instead of re-fetching the same external docs. Per-run isolated, external-only, success-only'
|
|
90
|
+
},
|
|
91
|
+
{
|
|
92
|
+
id: 'searchProvider',
|
|
93
|
+
label: 'search provider',
|
|
94
|
+
description: 'Engine behind web search (pi-worker-search + freshness checks). exa and ddg need no API key; brave needs BRAVE_SEARCH_API_KEY',
|
|
95
|
+
values: SEARCH_PROVIDERS
|
|
80
96
|
}
|
|
81
97
|
];
|
|
82
98
|
function makeTheme(theme) {
|
|
@@ -91,21 +107,32 @@ function makeTheme(theme) {
|
|
|
91
107
|
async function handleTaskConfig(_args, ctx) {
|
|
92
108
|
const cfg = { ...getConfig() };
|
|
93
109
|
if (ctx.mode !== 'tui') {
|
|
94
|
-
const lines = ITEMS.map(({ id, label }) => `${label.padEnd(22)} ${cfg[id]
|
|
110
|
+
const lines = ITEMS.map(({ id, label, values }) => `${label.padEnd(22)} ${values ? String(cfg[id])
|
|
111
|
+
: cfg[id] ? 'on'
|
|
112
|
+
: 'off'}`);
|
|
95
113
|
ctx.ui.notify(lines.join(' | '), 'info');
|
|
96
114
|
return;
|
|
97
115
|
}
|
|
98
116
|
await ctx.ui.custom((_tui, theme, _kb, done) => {
|
|
99
117
|
const listTheme = makeTheme(theme);
|
|
100
|
-
const items = ITEMS.map(({ id, label, description }) => ({
|
|
118
|
+
const items = ITEMS.map(({ id, label, description, values }) => ({
|
|
101
119
|
id,
|
|
102
120
|
label,
|
|
103
121
|
description,
|
|
104
|
-
currentValue: cfg[id]
|
|
105
|
-
|
|
122
|
+
currentValue: values ? String(cfg[id])
|
|
123
|
+
: cfg[id] ? 'on'
|
|
124
|
+
: 'off',
|
|
125
|
+
values: values ?? ['on', 'off']
|
|
106
126
|
}));
|
|
107
127
|
const list = new SettingsList(items, 10, listTheme, (id, newValue) => {
|
|
108
|
-
|
|
128
|
+
if (id === 'searchProvider') {
|
|
129
|
+
if (isSearchProvider(newValue))
|
|
130
|
+
cfg.searchProvider = newValue;
|
|
131
|
+
}
|
|
132
|
+
else {
|
|
133
|
+
;
|
|
134
|
+
cfg[id] = newValue === 'on';
|
|
135
|
+
}
|
|
109
136
|
saveConfig(cfg).catch(() => { });
|
|
110
137
|
}, () => done(undefined));
|
|
111
138
|
return new BorderedBox(list, CONFIG_TITLE, s => theme.fg('borderMuted', s), s => theme.fg('accent', theme.bold(s)));
|
|
@@ -0,0 +1,52 @@
|
|
|
1
|
+
/** One accepted-despite-FAIL record: the task and why its VERIFY failed. */
|
|
2
|
+
export interface AcceptDebt {
|
|
3
|
+
taskId: string;
|
|
4
|
+
reason: string;
|
|
5
|
+
}
|
|
6
|
+
export declare function acceptDebtFile(cwd: string): string;
|
|
7
|
+
/** The raw stored ledger ('' when none recorded yet). Parse with parseAcceptDebts. */
|
|
8
|
+
export declare function readAcceptDebtsRaw(cwd: string): Promise<string>;
|
|
9
|
+
/**
|
|
10
|
+
* Parse the stored ledger into records. A line without the separator (a reason but
|
|
11
|
+
* no id, e.g. hand-edited) parses with an empty taskId rather than being dropped —
|
|
12
|
+
* a recorded debt is never silently lost.
|
|
13
|
+
*/
|
|
14
|
+
export declare function parseAcceptDebts(raw: string): AcceptDebt[];
|
|
15
|
+
/** Read + parse in one step. */
|
|
16
|
+
export declare function readAcceptDebts(cwd: string): Promise<AcceptDebt[]>;
|
|
17
|
+
/**
|
|
18
|
+
* Append one accepted-despite-FAIL record, deduplicated against what is already
|
|
19
|
+
* stored (case-insensitive on task id + reason), keeping the newest MAX_DEBTS.
|
|
20
|
+
* Failures are swallowed — the ledger is an auditing aid, never a blocker of the
|
|
21
|
+
* gate sequence that calls it.
|
|
22
|
+
*/
|
|
23
|
+
export declare function recordAcceptDebt(cwd: string, taskId: string, reason: string): Promise<void>;
|
|
24
|
+
/** Overwrite the ledger with exactly these records (used to prune resolved debts). */
|
|
25
|
+
export declare function writeAcceptDebts(cwd: string, debts: AcceptDebt[]): Promise<void>;
|
|
26
|
+
/**
|
|
27
|
+
* STATIC-CLASS debt: one whose accepted FAIL was the deterministic whole-repo static
|
|
28
|
+
* health check (`repo health: …`, the prefix runWorkVerification's repoHealth branch
|
|
29
|
+
* emits). This is the ONE class a deterministic re-check can prove resolved
|
|
30
|
+
* stack-agnostically — the final gate runs the same static check, so a later task
|
|
31
|
+
* that fixed the statics resolves the debt. Every other reason is model-judged or
|
|
32
|
+
* behavioral and cannot be proven resolved without re-running the model.
|
|
33
|
+
*/
|
|
34
|
+
export declare function isStaticClassDebt(reason: string): boolean;
|
|
35
|
+
/**
|
|
36
|
+
* Re-check the ledger against the current run state. A static-class debt is RESOLVED
|
|
37
|
+
* iff the final gate's own static check now passes (`staticOk`); every other debt
|
|
38
|
+
* stays OPEN (unprovable ⇒ surface, never re-hide). FP-safe: the only auto-close is
|
|
39
|
+
* the one a deterministic check can stand behind.
|
|
40
|
+
*/
|
|
41
|
+
export declare function recheckAcceptDebts(debts: AcceptDebt[], opts: {
|
|
42
|
+
staticOk: boolean;
|
|
43
|
+
}): {
|
|
44
|
+
open: AcceptDebt[];
|
|
45
|
+
resolved: AcceptDebt[];
|
|
46
|
+
};
|
|
47
|
+
/**
|
|
48
|
+
* A one-line-per-debt suffix appended to the final gate's report reason so the still
|
|
49
|
+
* -open accepted defects surface in the gate outcome the user sees (and in the fail
|
|
50
|
+
* picker). Empty when nothing is open.
|
|
51
|
+
*/
|
|
52
|
+
export declare function buildAcceptDebtNote(open: AcceptDebt[]): string;
|
|
Binary file
|
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
import type { ExtensionAPI, ExtensionCommandContext } from '@earendil-works/pi-coding-agent';
|
|
2
2
|
import { type FinalGateFixFn } from './gate-deps.js';
|
|
3
3
|
import { type GateDeps } from './task-gates.js';
|
|
4
|
+
import type { AcceptDebt } from './accept-debt.js';
|
|
4
5
|
/**
|
|
5
6
|
* Injectable seams so the planner and loop are testable without spawning pi.
|
|
6
7
|
* `runChild` is the planning-only seam used by planAuto; everything else (runTask,
|
|
@@ -32,6 +33,7 @@ export interface AutoDeps extends GateDeps {
|
|
|
32
33
|
finalGate?: (cwd: string) => Promise<{
|
|
33
34
|
ok: boolean;
|
|
34
35
|
reason: string;
|
|
36
|
+
openDebts?: AcceptDebt[];
|
|
35
37
|
}>;
|
|
36
38
|
/**
|
|
37
39
|
* Bounded model-driven fix pass for a final-gate FAIL (see final-gate-fix.ts),
|
|
@@ -29,6 +29,7 @@ import { gitUnmergedPaths, gitStashRef } from './auto-commit.js';
|
|
|
29
29
|
import { runFinalIntegrationGate } from './final-gate.js';
|
|
30
30
|
import { classifyFinalGateAnswer, MAX_FINAL_GATE_AUTOFIX, FINAL_LEAVE_LABEL, FINAL_LEAVE_VALUE, FINAL_ACCEPT_LABEL, FINAL_ACCEPT_VALUE, FINAL_AUTOFIX_LABEL, FINAL_AUTOFIX_VALUE } from './final-gate-fix.js';
|
|
31
31
|
import { getConfig } from '../config/config.js';
|
|
32
|
+
import { configureResearchRun } from '../workers/research-cache.js';
|
|
32
33
|
import { CONTRACT_EXTRACT_PROMPT, parseContractLines, keepGroundedContracts, appendContracts } from './contracts.js';
|
|
33
34
|
// Hard ceiling on clarify questions per feature. The loop is open-ended (it stops
|
|
34
35
|
// when the model emits NONE), but a model that never says NONE would otherwise
|
|
@@ -681,6 +682,18 @@ export async function runAutoLoop(ctx, cwd, id, deps) {
|
|
|
681
682
|
let fin = await deps.finalGate(cwd);
|
|
682
683
|
if (!fin.ok)
|
|
683
684
|
await recGate(`final-gate: FAIL — ${fin.reason.slice(0, 300)}`);
|
|
685
|
+
// ACCEPT-debt re-check surfacing (mx5 run 4 B3 / run 8 TASK_0012):
|
|
686
|
+
// tasks the user accepted despite a verify-FAIL that the gate could
|
|
687
|
+
// not prove resolved against the current tree. Surface them at the
|
|
688
|
+
// gate moment — on PASS or FAIL — so a run never completes silently
|
|
689
|
+
// carrying an accepted defect. Informational: the per-task ACCEPT was
|
|
690
|
+
// already a human decision, so this reports, it does not re-fail.
|
|
691
|
+
if (fin.openDebts && fin.openDebts.length > 0) {
|
|
692
|
+
for (const d of fin.openDebts) {
|
|
693
|
+
await recGate(`accept-debt STILL OPEN — ${d.taskId || '(unknown task)'} was ACCEPTED despite verify-FAIL: ${d.reason.slice(0, 240)}`);
|
|
694
|
+
}
|
|
695
|
+
active.ui.notify(`${id}: ${fin.openDebts.length} task(s) accepted despite verify-FAIL are STILL unresolved at run end — see the gate trail.`, 'warning');
|
|
696
|
+
}
|
|
684
697
|
// Resolution loop: Leave-failed (recommended) / Autofix (bounded,
|
|
685
698
|
// model-driven fix pass + gate re-run — run 7's gap: the picker
|
|
686
699
|
// had NO automated fix path) / Accept. The user always decides;
|
|
@@ -916,6 +929,10 @@ async function handleTaskAuto(args, ctx) {
|
|
|
916
929
|
return;
|
|
917
930
|
}
|
|
918
931
|
autoRunning = true;
|
|
932
|
+
// Stamp a fresh per-run research-cache id (F10) BEFORE planning so enrichment and
|
|
933
|
+
// every task's research phase share one run's cache; disabled ⇒ clears any token a
|
|
934
|
+
// prior run left, so nothing is cached.
|
|
935
|
+
configureResearchRun(getConfig().researchCache);
|
|
919
936
|
const abort = new AbortController();
|
|
920
937
|
const deps = defaultDeps(ctx, cwd, abort.signal, deriveTitle(raw));
|
|
921
938
|
let id;
|
|
@@ -958,6 +975,9 @@ async function handleTaskAutoResume(_args, ctx) {
|
|
|
958
975
|
ctx.ui.notify(`Resuming ${id}…`, 'info');
|
|
959
976
|
await updateTaskFrontMatter(cwd, id, { state: 'in_progress' });
|
|
960
977
|
autoRunning = true;
|
|
978
|
+
// Fresh per-run research-cache id for the resumed run (F10); a resume re-fetches
|
|
979
|
+
// rather than reusing the interrupted run's digest — safe, only slightly less reuse.
|
|
980
|
+
configureResearchRun(getConfig().researchCache);
|
|
961
981
|
const abort = new AbortController();
|
|
962
982
|
// Resume only runs the loop (runTask); no planning children, so the loader
|
|
963
983
|
// title is unused here — pass the id for clarity if that ever changes.
|
|
@@ -1,9 +1,17 @@
|
|
|
1
1
|
import { type HealthCommand } from './repo-health-check.js';
|
|
2
|
+
import { type AcceptDebt } from './accept-debt.js';
|
|
2
3
|
export interface FinalGateOutcome {
|
|
3
4
|
/** true → statics and every runnable integration command passed (or nothing to run). */
|
|
4
5
|
ok: boolean;
|
|
5
6
|
/** On a fail: the exact command, its exit code, and the tail of its output. */
|
|
6
7
|
reason: string;
|
|
8
|
+
/**
|
|
9
|
+
* ACCEPT-despite-verify-FAIL debts still open at run end (mx5 run 4 B3 / run 8
|
|
10
|
+
* TASK_0012): tasks the user blessed as-is despite a verify-FAIL that a
|
|
11
|
+
* deterministic re-check could not prove resolved. The caller surfaces them so a
|
|
12
|
+
* run never completes silently carrying an accepted defect. Empty/absent = none.
|
|
13
|
+
*/
|
|
14
|
+
openDebts?: AcceptDebt[];
|
|
7
15
|
}
|
|
8
16
|
/**
|
|
9
17
|
* The project's OWN whole-repo integration commands (test, then build — test
|
package/dist/task/final-gate.js
CHANGED
|
@@ -42,6 +42,7 @@ import { spawn, spawnSync } from 'node:child_process';
|
|
|
42
42
|
import { existsSync, readFileSync } from 'node:fs';
|
|
43
43
|
import * as path from 'node:path';
|
|
44
44
|
import { runRepoHealthCheck, discoverHealthCommands } from './repo-health-check.js';
|
|
45
|
+
import { readAcceptDebts, recheckAcceptDebts, writeAcceptDebts, buildAcceptDebtNote } from './accept-debt.js';
|
|
45
46
|
function packageScripts(cwd) {
|
|
46
47
|
try {
|
|
47
48
|
const j = JSON.parse(readFileSync(path.join(cwd, 'package.json'), 'utf8'));
|
|
@@ -280,13 +281,32 @@ function runGateCommand(cwd, [bin, args], timeoutMs) {
|
|
|
280
281
|
*/
|
|
281
282
|
export async function runFinalIntegrationGate(cwd, timeoutMs = 900_000, bootGraceMs = 10_000) {
|
|
282
283
|
const stat = runRepoHealthCheck(cwd);
|
|
284
|
+
// ACCEPT-debt re-check (mx5 run 4 B3 / run 8 TASK_0012): read the ledger of tasks
|
|
285
|
+
// the user accepted despite a verify-FAIL and re-check each against the current
|
|
286
|
+
// tree. A static-class debt whose statics now pass is provably RESOLVED (a later
|
|
287
|
+
// task fixed it) and pruned; every other debt cannot be proven resolved
|
|
288
|
+
// deterministically, so it stays OPEN and is surfaced in this gate's report — a
|
|
289
|
+
// run may not complete silently carrying an accepted defect. FP-safe by
|
|
290
|
+
// construction (see accept-debt.ts). Best-effort: a ledger read/write failure
|
|
291
|
+
// must never break the gate.
|
|
292
|
+
const { open: openDebts, resolved } = recheckAcceptDebts(await readAcceptDebts(cwd), {
|
|
293
|
+
staticOk: stat.ok
|
|
294
|
+
});
|
|
295
|
+
if (resolved.length > 0)
|
|
296
|
+
await writeAcceptDebts(cwd, openDebts);
|
|
297
|
+
const debtNote = buildAcceptDebtNote(openDebts);
|
|
298
|
+
const withDebts = (o) => ({
|
|
299
|
+
...o,
|
|
300
|
+
reason: `${o.reason}${debtNote}`,
|
|
301
|
+
openDebts
|
|
302
|
+
});
|
|
283
303
|
if (!stat.ok)
|
|
284
|
-
return { ok: false, reason: `static checks: ${stat.reason}` };
|
|
304
|
+
return withDebts({ ok: false, reason: `static checks: ${stat.reason}` });
|
|
285
305
|
const lockCmds = discoverLockfileChecks(cwd);
|
|
286
306
|
const { cmds } = discoverIntegrationCommands(cwd);
|
|
287
307
|
const boot = discoverBootCommand(cwd);
|
|
288
308
|
if (lockCmds.length === 0 && cmds.length === 0 && !boot) {
|
|
289
|
-
return { ok: true, reason: 'no integration command found (statics passed)' };
|
|
309
|
+
return withDebts({ ok: true, reason: 'no integration command found (statics passed)' });
|
|
290
310
|
}
|
|
291
311
|
const ran = [];
|
|
292
312
|
for (const { prefix, list } of [
|
|
@@ -299,10 +319,10 @@ export async function runFinalIntegrationGate(cwd, timeoutMs = 900_000, bootGrac
|
|
|
299
319
|
if (r.outcome === 'skip')
|
|
300
320
|
continue;
|
|
301
321
|
if (r.outcome === 'fail') {
|
|
302
|
-
return {
|
|
322
|
+
return withDebts({
|
|
303
323
|
ok: false,
|
|
304
324
|
reason: `${prefix}\`${label}\` exited ${r.status}${r.tail ? ` — ${r.tail}` : ''}`
|
|
305
|
-
};
|
|
325
|
+
});
|
|
306
326
|
}
|
|
307
327
|
ran.push(label);
|
|
308
328
|
}
|
|
@@ -311,15 +331,15 @@ export async function runFinalIntegrationGate(cwd, timeoutMs = 900_000, bootGrac
|
|
|
311
331
|
const label = `${boot[0]} ${boot[1].join(' ')}`;
|
|
312
332
|
const b = await runBootCheck(cwd, boot, bootGraceMs);
|
|
313
333
|
if (b.outcome === 'fail') {
|
|
314
|
-
return { ok: false, reason: `boot check: \`${label}\` ${b.detail}` };
|
|
334
|
+
return withDebts({ ok: false, reason: `boot check: \`${label}\` ${b.detail}` });
|
|
315
335
|
}
|
|
316
336
|
if (b.outcome === 'pass')
|
|
317
337
|
ran.push(label);
|
|
318
338
|
}
|
|
319
|
-
return {
|
|
339
|
+
return withDebts({
|
|
320
340
|
ok: true,
|
|
321
341
|
reason: ran.length > 0 ?
|
|
322
342
|
`statics + ${ran.map(c => `\`${c}\``).join(', ')} passed`
|
|
323
343
|
: 'statics passed (integration commands not runnable here)'
|
|
324
|
-
};
|
|
344
|
+
});
|
|
325
345
|
}
|
|
@@ -0,0 +1,39 @@
|
|
|
1
|
+
/** Run a git subcommand in the guard's cwd; only stdout + exit code are read. */
|
|
2
|
+
export type FrozenGit = (args: string[]) => Promise<{
|
|
3
|
+
stdout: string;
|
|
4
|
+
exitCode: number;
|
|
5
|
+
}>;
|
|
6
|
+
/**
|
|
7
|
+
* The concrete paths the spec forbids modifying, normalized and de-duplicated —
|
|
8
|
+
* the SAME extraction the verify prohibition-probe and the accept-debt ledger
|
|
9
|
+
* consume, so "frozen" means one thing across the gates. Empty when the spec is
|
|
10
|
+
* null or names no path-like token on a modification-ban line: the guard is then a
|
|
11
|
+
* no-op by construction.
|
|
12
|
+
*/
|
|
13
|
+
export declare function frozenPathsFromSpec(spec: string | null | undefined): string[];
|
|
14
|
+
/**
|
|
15
|
+
* Parse `git status --porcelain` output (already scoped to the frozen pathspec)
|
|
16
|
+
* into the list of changed files, for the gate-trail record and to decide whether
|
|
17
|
+
* anything must be reverted at all. A rename line (`R old -> new`) yields the NEW
|
|
18
|
+
* path — the side that carries the child's write. Deterministic and pure so the
|
|
19
|
+
* parsing is unit-tested without a real repo.
|
|
20
|
+
*/
|
|
21
|
+
export declare function parseChangedFrozenFiles(porcelain: string): string[];
|
|
22
|
+
/**
|
|
23
|
+
* Restore the spec-frozen paths to their committed (HEAD) state, undoing any
|
|
24
|
+
* change a just-run write-capable gate child made to them, and return the list of
|
|
25
|
+
* files that had to be reverted (empty ⇒ the pass respected every frozen path).
|
|
26
|
+
*
|
|
27
|
+
* Runs AFTER the task's own work is committed (HEAD), so "restore to HEAD" keeps
|
|
28
|
+
* the verified task's version of the frozen file and discards ONLY the gate
|
|
29
|
+
* child's edit on top of it — the task's own frozen-path edits, if any, are a
|
|
30
|
+
* separate concern the verify prohibition-probe surfaces. `git checkout HEAD`
|
|
31
|
+
* covers modified/deleted tracked files under each pathspec; `git clean` removes
|
|
32
|
+
* any untracked file the pass created under a frozen directory. Both are scoped to
|
|
33
|
+
* the frozen pathspec, so nothing else in the tree is touched.
|
|
34
|
+
*
|
|
35
|
+
* Best-effort: an empty frozen list, a non-git tree, or any git error yields an
|
|
36
|
+
* empty result — the guard must never break the gate on a project it cannot reason
|
|
37
|
+
* about.
|
|
38
|
+
*/
|
|
39
|
+
export declare function revertFrozenPaths(paths: string[], git: FrozenGit): Promise<string[]>;
|
|
@@ -0,0 +1,116 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* frozen-path-guard — deterministic write-deny on spec-frozen paths for the
|
|
3
|
+
* write-capable gate children (the enforce EDIT pass especially).
|
|
4
|
+
*
|
|
5
|
+
* The failure class (carried from mx5 run 6, deferred there as the last unshipped
|
|
6
|
+
* piece of the frozen-contract work): a spec's CONSTRAINTS pin a path as
|
|
7
|
+
* off-limits ("**Do NOT modify** `src/server/index.ts`"), and a later
|
|
8
|
+
* write-enabled GATE pass — the enforce child runs `read,edit` — edits that path
|
|
9
|
+
* anyway. Its edits are judged only locally (a guideline-compliance verdict), then
|
|
10
|
+
* committed as an "ENFORCE GUIDELINES" snapshot on top of the verified task. The
|
|
11
|
+
* frozen contract is silently mutated by the very pass meant to police the work.
|
|
12
|
+
*
|
|
13
|
+
* Prompt framing is A/B-PROVEN insufficient for this class (the "FROZEN CONTRACT,
|
|
14
|
+
* MUST NOT edit" instruction held 0/5 unguarded, ~1/5 with framing — the weak
|
|
15
|
+
* local model ignores an explicit capitalized MUST-NOT most of the time). A
|
|
16
|
+
* chmod-style physical deny holds 5/5 but makes the model THRASH on the bare
|
|
17
|
+
* EACCES (1000+ futile re-attempts observed) — unacceptable on the enforce child,
|
|
18
|
+
* which runs UNGUARDED (no wall-clock timeout). pi itself has no path-level edit
|
|
19
|
+
* interception (only tool-NAME allow/deny), so the achievable, thrash-free,
|
|
20
|
+
* stack-agnostic realization of a tool-layer deny is this: let the pass edit
|
|
21
|
+
* freely, then deterministically UNDO any frozen-path change it made before those
|
|
22
|
+
* edits can be committed. The violating write never lands in a commit, regardless
|
|
23
|
+
* of what the model intended — no model in the loop, same discipline as
|
|
24
|
+
* git-state-guard.
|
|
25
|
+
*
|
|
26
|
+
* Everything degrades to a no-op when the spec froze nothing (a single-`/task`
|
|
27
|
+
* run, or a spec with no `Do NOT modify` line naming a path → no frozen paths →
|
|
28
|
+
* nothing snapshotted, nothing reverted) and on any git error. Pure git shape,
|
|
29
|
+
* zero stack assumptions — a frozen path exists for a CLI, a library, a script
|
|
30
|
+
* collection exactly as for a web app.
|
|
31
|
+
*/
|
|
32
|
+
import { extractProhibitions } from './prohibition-probe.js';
|
|
33
|
+
/**
|
|
34
|
+
* The concrete paths the spec forbids modifying, normalized and de-duplicated —
|
|
35
|
+
* the SAME extraction the verify prohibition-probe and the accept-debt ledger
|
|
36
|
+
* consume, so "frozen" means one thing across the gates. Empty when the spec is
|
|
37
|
+
* null or names no path-like token on a modification-ban line: the guard is then a
|
|
38
|
+
* no-op by construction.
|
|
39
|
+
*/
|
|
40
|
+
export function frozenPathsFromSpec(spec) {
|
|
41
|
+
if (!spec)
|
|
42
|
+
return [];
|
|
43
|
+
const seen = new Set();
|
|
44
|
+
for (const p of extractProhibitions(spec)) {
|
|
45
|
+
const n = p.path.replace(/^\.\//, '').replace(/\/+$/, '');
|
|
46
|
+
if (n.length > 0)
|
|
47
|
+
seen.add(n);
|
|
48
|
+
}
|
|
49
|
+
return [...seen];
|
|
50
|
+
}
|
|
51
|
+
/**
|
|
52
|
+
* Parse `git status --porcelain` output (already scoped to the frozen pathspec)
|
|
53
|
+
* into the list of changed files, for the gate-trail record and to decide whether
|
|
54
|
+
* anything must be reverted at all. A rename line (`R old -> new`) yields the NEW
|
|
55
|
+
* path — the side that carries the child's write. Deterministic and pure so the
|
|
56
|
+
* parsing is unit-tested without a real repo.
|
|
57
|
+
*/
|
|
58
|
+
export function parseChangedFrozenFiles(porcelain) {
|
|
59
|
+
const out = [];
|
|
60
|
+
const seen = new Set();
|
|
61
|
+
for (const raw of porcelain.split('\n')) {
|
|
62
|
+
// Porcelain v1: two status chars, a space, then the path. Blank/short lines
|
|
63
|
+
// (trailing newline) carry no entry.
|
|
64
|
+
if (raw.length < 4)
|
|
65
|
+
continue;
|
|
66
|
+
let file = raw.slice(3).trim();
|
|
67
|
+
if (file.length === 0)
|
|
68
|
+
continue;
|
|
69
|
+
// Rename/copy: "orig -> new" — the new path is the one the write produced.
|
|
70
|
+
const arrow = file.indexOf(' -> ');
|
|
71
|
+
if (arrow !== -1)
|
|
72
|
+
file = file.slice(arrow + 4).trim();
|
|
73
|
+
// Porcelain quotes paths with unusual chars; strip the surrounding quotes.
|
|
74
|
+
if (file.startsWith('"') && file.endsWith('"') && file.length >= 2) {
|
|
75
|
+
file = file.slice(1, -1);
|
|
76
|
+
}
|
|
77
|
+
if (file.length === 0 || seen.has(file))
|
|
78
|
+
continue;
|
|
79
|
+
seen.add(file);
|
|
80
|
+
out.push(file);
|
|
81
|
+
}
|
|
82
|
+
return out;
|
|
83
|
+
}
|
|
84
|
+
/**
|
|
85
|
+
* Restore the spec-frozen paths to their committed (HEAD) state, undoing any
|
|
86
|
+
* change a just-run write-capable gate child made to them, and return the list of
|
|
87
|
+
* files that had to be reverted (empty ⇒ the pass respected every frozen path).
|
|
88
|
+
*
|
|
89
|
+
* Runs AFTER the task's own work is committed (HEAD), so "restore to HEAD" keeps
|
|
90
|
+
* the verified task's version of the frozen file and discards ONLY the gate
|
|
91
|
+
* child's edit on top of it — the task's own frozen-path edits, if any, are a
|
|
92
|
+
* separate concern the verify prohibition-probe surfaces. `git checkout HEAD`
|
|
93
|
+
* covers modified/deleted tracked files under each pathspec; `git clean` removes
|
|
94
|
+
* any untracked file the pass created under a frozen directory. Both are scoped to
|
|
95
|
+
* the frozen pathspec, so nothing else in the tree is touched.
|
|
96
|
+
*
|
|
97
|
+
* Best-effort: an empty frozen list, a non-git tree, or any git error yields an
|
|
98
|
+
* empty result — the guard must never break the gate on a project it cannot reason
|
|
99
|
+
* about.
|
|
100
|
+
*/
|
|
101
|
+
export async function revertFrozenPaths(paths, git) {
|
|
102
|
+
if (paths.length === 0)
|
|
103
|
+
return [];
|
|
104
|
+
const status = await git(['status', '--porcelain', '--', ...paths]);
|
|
105
|
+
if (status.exitCode !== 0)
|
|
106
|
+
return [];
|
|
107
|
+
const changed = parseChangedFrozenFiles(status.stdout);
|
|
108
|
+
if (changed.length === 0)
|
|
109
|
+
return [];
|
|
110
|
+
// Restore tracked modifications/deletions from HEAD, then remove any untracked
|
|
111
|
+
// additions — both confined to the frozen pathspec so the pass's legitimate
|
|
112
|
+
// edits to OTHER files survive untouched.
|
|
113
|
+
await git(['checkout', '-f', 'HEAD', '--', ...paths]);
|
|
114
|
+
await git(['clean', '-fdq', '--', ...paths]);
|
|
115
|
+
return changed;
|
|
116
|
+
}
|
package/dist/task/gate-deps.js
CHANGED
|
@@ -21,11 +21,13 @@ import { runGuidelineEnforcement, classifyEnforceChildFailure } from './enforce-
|
|
|
21
21
|
import { runWorkVerification, extractSpecForVerification } from './verify-work.js';
|
|
22
22
|
import { readEnvNotes, appendEnvNotes } from './env-notes.js';
|
|
23
23
|
import { readContracts } from './contracts.js';
|
|
24
|
+
import { recordAcceptDebt } from './accept-debt.js';
|
|
24
25
|
import { runRepoHealthCheck } from './repo-health-check.js';
|
|
25
26
|
import { runFinalIntegrationGate, discoverGateCommandLabels } from './final-gate.js';
|
|
26
27
|
import { runFinalGateAutofix } from './final-gate-fix.js';
|
|
27
28
|
import { researchResolution } from './verify-resolution.js';
|
|
28
29
|
import { extractProhibitions, findProhibitionViolations } from './prohibition-probe.js';
|
|
30
|
+
import { frozenPathsFromSpec, revertFrozenPaths } from './frozen-path-guard.js';
|
|
29
31
|
import { findProbeGaming, parseAddedLines } from './probe-gaming.js';
|
|
30
32
|
import { findSubstitutionSuspects, isTestFile } from './substitution-probe.js';
|
|
31
33
|
import { findTestRebuiltAssemblies, testAssemblyVerifyFindings } from './test-assembly.js';
|
|
@@ -264,6 +266,29 @@ export function buildGateDeps(params) {
|
|
|
264
266
|
// Durable per-task gate trail: every verdict/decision lands in the task
|
|
265
267
|
// file's `## gates` section so gate behavior is auditable from artifacts.
|
|
266
268
|
record: (cwd2, taskId, line) => appendGateRecord(cwd2, taskId, line),
|
|
269
|
+
// Durable ACCEPT-despite-verify-FAIL ledger under .pi-tasks/ (survives
|
|
270
|
+
// discardEdits): the final integration gate re-checks each debt at run end.
|
|
271
|
+
recordAcceptDebt: (cwd2, taskId, reason) => recordAcceptDebt(cwd2, taskId, reason),
|
|
272
|
+
// Frozen-path write-deny (see frozen-path-guard.ts): the concrete paths this
|
|
273
|
+
// task's spec forbids modifying, so the gate sequence can UNDO any edit the
|
|
274
|
+
// enforce EDIT pass makes to them before those edits are committed. Reads the
|
|
275
|
+
// same composed spec + extractProhibitions the verify probe consumes; empty on
|
|
276
|
+
// a spec that froze nothing → the guard is a no-op.
|
|
277
|
+
frozenPaths: async (cwd2, taskId) => {
|
|
278
|
+
try {
|
|
279
|
+
const { body } = await readTaskFile(cwd2, taskId);
|
|
280
|
+
return frozenPathsFromSpec(extractSpecForVerification(body));
|
|
281
|
+
}
|
|
282
|
+
catch {
|
|
283
|
+
return [];
|
|
284
|
+
}
|
|
285
|
+
},
|
|
286
|
+
// Restore those frozen paths to HEAD, discarding a gate child's edits to them;
|
|
287
|
+
// returns the files actually reverted (for the trail). Best-effort git shape.
|
|
288
|
+
revertFrozenPaths: (cwd2, paths) => revertFrozenPaths(paths, async (args) => {
|
|
289
|
+
const r = await git(cwd2, args, signal);
|
|
290
|
+
return { stdout: r.stdout, exitCode: r.exitCode };
|
|
291
|
+
}),
|
|
267
292
|
commit: (cwd2, message) => getConfig().autoCommit ?
|
|
268
293
|
gitCommitAll(cwd2, message, signal)
|
|
269
294
|
: Promise.resolve({ committed: false, reason: 'auto-commit disabled' }),
|
package/dist/task/phases.d.ts
CHANGED
|
@@ -6,6 +6,7 @@ import type { ExtensionCommandContext } from '@earendil-works/pi-coding-agent';
|
|
|
6
6
|
import { docsFocused } from '../workers/docs-core.js';
|
|
7
7
|
import { fetchFocused } from '../workers/fetch-core.js';
|
|
8
8
|
import type { SearchCoreInput, SearchCoreResult } from '../workers/search-core.js';
|
|
9
|
+
import type { SearchProvider } from '../workers/search-types.js';
|
|
9
10
|
import { type ExternalContextDeps } from './external-context.js';
|
|
10
11
|
import { MAX_GRILL_QUESTIONS } from './prompts.js';
|
|
11
12
|
import { type PhaseName } from './task-types.js';
|
|
@@ -63,8 +64,11 @@ export declare function phaseVerifyTooling(deps: PhaseDeps, research: string): P
|
|
|
63
64
|
export interface PhaseResearchDeps extends ExternalContextDeps {
|
|
64
65
|
getFileInventory?: (cwd: string, signal?: AbortSignal) => Promise<string>;
|
|
65
66
|
}
|
|
66
|
-
/**
|
|
67
|
-
|
|
67
|
+
/**
|
|
68
|
+
* Is live web search configured for this process? The keyless providers (exa,
|
|
69
|
+
* ddg) always are; only brave needs its API key — mirrors search-core's lookup.
|
|
70
|
+
*/
|
|
71
|
+
export declare function searchConfigured(getEnv?: (k: string) => string | undefined, provider?: SearchProvider): boolean;
|
|
68
72
|
/** Extra prompt block for the APIS worker when search is available — trigger-framed
|
|
69
73
|
* (the validated shape for getting a local model to actually reach for search). */
|
|
70
74
|
export declare const RESEARCH_SEARCH_HINT: string;
|
package/dist/task/phases.js
CHANGED
|
@@ -166,8 +166,13 @@ const DOCS_EXTENSION_PATH = new URL('../workers/docs-extension.js', import.meta.
|
|
|
166
166
|
* STRUCTURAL: three consecutive audited runs made 0 search calls because the
|
|
167
167
|
* child literally did not have the tool. */
|
|
168
168
|
const SEARCH_EXTENSION_PATH = new URL('../workers/search-extension.js', import.meta.url).pathname;
|
|
169
|
-
/**
|
|
170
|
-
|
|
169
|
+
/**
|
|
170
|
+
* Is live web search configured for this process? The keyless providers (exa,
|
|
171
|
+
* ddg) always are; only brave needs its API key — mirrors search-core's lookup.
|
|
172
|
+
*/
|
|
173
|
+
export function searchConfigured(getEnv = k => process.env[k], provider = getConfig().searchProvider) {
|
|
174
|
+
if (provider !== 'brave')
|
|
175
|
+
return true;
|
|
171
176
|
return Boolean(getEnv('BRAVE_SEARCH_API_KEY') ?? getEnv('BRAVE_API_KEY'));
|
|
172
177
|
}
|
|
173
178
|
/** Extra prompt block for the APIS worker when search is available — trigger-framed
|
|
@@ -6,7 +6,18 @@ export interface HealthOutcome {
|
|
|
6
6
|
reason: string;
|
|
7
7
|
/** Which manifest drove discovery, or null when none was found. */
|
|
8
8
|
ecosystem: string | null;
|
|
9
|
+
/**
|
|
10
|
+
* First lines of the failing command's combined stderr+stdout — captured so a
|
|
11
|
+
* FAIL is explainable from artifacts alone. Run-8 F8: five enforce passes were
|
|
12
|
+
* discarded on "`bun run lint` exited 2" and the cause was unreproducible
|
|
13
|
+
* post-run because only the exit code was recorded (exit 2 is the linter's
|
|
14
|
+
* CRASH class; findings exit 1 — the captured output is what tells them apart).
|
|
15
|
+
* Empty string on pass / skip.
|
|
16
|
+
*/
|
|
17
|
+
output: string;
|
|
9
18
|
}
|
|
19
|
+
/** Combine a failing command's stderr+stdout into a bounded, first-N-lines snippet. */
|
|
20
|
+
export declare function captureHealthOutput(stdout: string, stderr: string): string;
|
|
10
21
|
/** One discovered command: the binary and its args, run from the repo root. */
|
|
11
22
|
export type HealthCommand = [bin: string, args: string[]];
|
|
12
23
|
/**
|