@miller-tech/uap 1.128.0 → 1.128.2
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/.tsbuildinfo +1 -1
- package/dist/cli/deliver.js +23 -4
- package/dist/cli/deliver.js.map +1 -1
- package/dist/delivery/snapshot.d.ts +45 -5
- package/dist/delivery/snapshot.d.ts.map +1 -1
- package/dist/delivery/snapshot.js +304 -21
- package/dist/delivery/snapshot.js.map +1 -1
- package/docs/guides/DELIVER.md +13 -0
- package/docs/reference/CONFIGURATION.md +2 -0
- package/package.json +2 -2
- package/src/policies/enforcers/__pycache__/_common.cpython-312.pyc +0 -0
- package/src/policies/enforcers/delivery_enforcement.py +10 -1
- package/src/policies/enforcers/schema_diff_gate.py +25 -8
- package/tools/agents/scripts/__pycache__/toolcall_path_normalizer.cpython-312.pyc +0 -0
- package/tools/agents/tests/test_schema_diff_gate.py +115 -0
|
@@ -3,31 +3,314 @@
|
|
|
3
3
|
*
|
|
4
4
|
* Captures the project state before the convergence loop runs so deliver can
|
|
5
5
|
* roll back if it ends up worse (by the real gates' measure) than it started.
|
|
6
|
-
* Excludes heavy/derived
|
|
7
|
-
* task's source and would make snapshots slow.
|
|
6
|
+
* Excludes heavy/derived DIRECTORIES (.git, node_modules, target, .venv, …):
|
|
7
|
+
* they aren't part of the task's source and would make snapshots slow. The
|
|
8
|
+
* contract is symmetric — an excluded directory is neither snapshotted nor
|
|
9
|
+
* touched by restore, at any depth. Files sharing an excluded name (e.g. a
|
|
10
|
+
* script called `build`) are snapshotted normally.
|
|
11
|
+
*
|
|
12
|
+
* Hardening (2026-07-08, after the project-i incident where a 424 GB Rust
|
|
13
|
+
* `target/` tree was copied into a 61 GB RAM tmpfs):
|
|
14
|
+
* - snapshots land on real disk (~/.cache/uap/snapshots, UAP_SNAPSHOT_DIR
|
|
15
|
+
* overrides — absolute paths only), never os.tmpdir(), which is RAM-backed
|
|
16
|
+
* on many Linux setups;
|
|
17
|
+
* - the source tree is size-guarded before copying (UAP_SNAPSHOT_MAX_MB,
|
|
18
|
+
* default 4096) — over the cap, snapshotTree returns null and the caller
|
|
19
|
+
* degrades to "no rollback this run" instead of exhausting the machine;
|
|
20
|
+
* - snapshotTree never throws: any failure cleans up its partial copy and
|
|
21
|
+
* returns null;
|
|
22
|
+
* - restoreTree validates the snapshot before touching the project, restores
|
|
23
|
+
* copy-first (prune extras, then merge back) so an interrupted restore
|
|
24
|
+
* never leaves a gutted tree, and marks the snapshot preserve-on-failure
|
|
25
|
+
* so the reaper won't collect the only good copy;
|
|
26
|
+
* - each snapshot carries a pid marker; stale snapshots from dead processes
|
|
27
|
+
* (crashes, SIGKILL, ENOSPC mid-copy) are reaped before a new one is taken.
|
|
28
|
+
*/
|
|
29
|
+
import { cpSync, rmSync, mkdtempSync, mkdirSync, readdirSync, readFileSync, writeFileSync, lstatSync, statSync, } from 'fs';
|
|
30
|
+
import { join, basename, isAbsolute, resolve } from 'path';
|
|
31
|
+
import { tmpdir, homedir } from 'os';
|
|
32
|
+
const EXCLUDE = new Set([
|
|
33
|
+
// VCS / UAP-internal
|
|
34
|
+
'.git',
|
|
35
|
+
'.worktrees',
|
|
36
|
+
'.uap-deliver',
|
|
37
|
+
'.uap-backups',
|
|
38
|
+
// JS
|
|
39
|
+
'node_modules',
|
|
40
|
+
'dist',
|
|
41
|
+
'build',
|
|
42
|
+
'out',
|
|
43
|
+
'coverage',
|
|
44
|
+
'.next',
|
|
45
|
+
'.nuxt',
|
|
46
|
+
'.svelte-kit',
|
|
47
|
+
'.turbo',
|
|
48
|
+
'.cache',
|
|
49
|
+
'.parcel-cache',
|
|
50
|
+
// Rust / JVM / Go
|
|
51
|
+
'target',
|
|
52
|
+
'.gradle',
|
|
53
|
+
'vendor',
|
|
54
|
+
// Python
|
|
55
|
+
'.venv',
|
|
56
|
+
'venv',
|
|
57
|
+
'__pycache__',
|
|
58
|
+
'.pytest_cache',
|
|
59
|
+
'.mypy_cache',
|
|
60
|
+
'.ruff_cache',
|
|
61
|
+
'.tox',
|
|
62
|
+
]);
|
|
63
|
+
/** Marker written inside each snapshot so the reaper can tell live from stale. */
|
|
64
|
+
const META_FILE = '.uap-snap.json';
|
|
65
|
+
const SNAP_PREFIX = 'uap-snap-';
|
|
66
|
+
/** Age past which a marker-less (legacy/partial) snapshot is considered stale. */
|
|
67
|
+
const LEGACY_STALE_MS = 24 * 60 * 60 * 1000;
|
|
68
|
+
/** Backstop against pid reuse pinning a dead snapshot forever. */
|
|
69
|
+
const MAX_AGE_MS = 7 * 24 * 60 * 60 * 1000;
|
|
70
|
+
const DEFAULT_MAX_MB = 4096;
|
|
71
|
+
/** True for a directory whose name marks it as derived/heavy (dir-only contract). */
|
|
72
|
+
function isExcludedDir(path, name) {
|
|
73
|
+
if (!EXCLUDE.has(name))
|
|
74
|
+
return false;
|
|
75
|
+
try {
|
|
76
|
+
return lstatSync(path).isDirectory();
|
|
77
|
+
}
|
|
78
|
+
catch {
|
|
79
|
+
return false;
|
|
80
|
+
}
|
|
81
|
+
}
|
|
82
|
+
/** Disk-backed base directory for snapshots (never os.tmpdir — often RAM tmpfs). */
|
|
83
|
+
export function snapshotBaseDir() {
|
|
84
|
+
const raw = process.env.UAP_SNAPSHOT_DIR?.trim();
|
|
85
|
+
// Relative/garbage overrides silently landing inside the project tree would
|
|
86
|
+
// snapshot into the thing being snapshotted — absolute paths only.
|
|
87
|
+
const base = raw && isAbsolute(raw) ? resolve(raw) : join(homedir(), '.cache', 'uap', 'snapshots');
|
|
88
|
+
mkdirSync(base, { recursive: true });
|
|
89
|
+
return base;
|
|
90
|
+
}
|
|
91
|
+
function maxSnapshotBytes() {
|
|
92
|
+
const mb = Number(process.env.UAP_SNAPSHOT_MAX_MB);
|
|
93
|
+
return (Number.isFinite(mb) && mb > 0 ? mb : DEFAULT_MAX_MB) * 1024 * 1024;
|
|
94
|
+
}
|
|
95
|
+
/**
|
|
96
|
+
* Sum file sizes under root (skipping excluded dirs and symlinks), bailing out
|
|
97
|
+
* as soon as the running total exceeds `limit` so huge trees cost one early
|
|
98
|
+
* partial walk, not a full traversal.
|
|
99
|
+
*/
|
|
100
|
+
function treeSizeExceeds(root, limit) {
|
|
101
|
+
let total = 0;
|
|
102
|
+
const stack = [root];
|
|
103
|
+
while (stack.length > 0) {
|
|
104
|
+
const dir = stack.pop();
|
|
105
|
+
let entries;
|
|
106
|
+
try {
|
|
107
|
+
entries = readdirSync(dir);
|
|
108
|
+
}
|
|
109
|
+
catch {
|
|
110
|
+
continue; // unreadable dir — the copy will surface real problems
|
|
111
|
+
}
|
|
112
|
+
for (const name of entries) {
|
|
113
|
+
const path = join(dir, name);
|
|
114
|
+
let st;
|
|
115
|
+
try {
|
|
116
|
+
st = lstatSync(path);
|
|
117
|
+
}
|
|
118
|
+
catch {
|
|
119
|
+
continue;
|
|
120
|
+
}
|
|
121
|
+
if (st.isSymbolicLink())
|
|
122
|
+
continue;
|
|
123
|
+
if (st.isDirectory()) {
|
|
124
|
+
if (!EXCLUDE.has(name))
|
|
125
|
+
stack.push(path);
|
|
126
|
+
}
|
|
127
|
+
else {
|
|
128
|
+
total += st.size;
|
|
129
|
+
if (total > limit)
|
|
130
|
+
return true;
|
|
131
|
+
}
|
|
132
|
+
}
|
|
133
|
+
}
|
|
134
|
+
return false;
|
|
135
|
+
}
|
|
136
|
+
function pidAlive(pid) {
|
|
137
|
+
try {
|
|
138
|
+
process.kill(pid, 0);
|
|
139
|
+
return true;
|
|
140
|
+
}
|
|
141
|
+
catch (err) {
|
|
142
|
+
// EPERM means the process exists but isn't ours.
|
|
143
|
+
return err.code === 'EPERM';
|
|
144
|
+
}
|
|
145
|
+
}
|
|
146
|
+
function isStaleSnapshot(path) {
|
|
147
|
+
try {
|
|
148
|
+
const metaPath = join(path, META_FILE);
|
|
149
|
+
// The tmpdir sweep reads world-writable locations: a planted FIFO or
|
|
150
|
+
// symlink-to-/dev/zero marker must not hang or balloon the read.
|
|
151
|
+
const st = lstatSync(metaPath);
|
|
152
|
+
if (!st.isFile() || st.size > 4096)
|
|
153
|
+
throw new Error('untrusted marker');
|
|
154
|
+
const meta = JSON.parse(readFileSync(metaPath, 'utf-8'));
|
|
155
|
+
if (meta.preserve === true)
|
|
156
|
+
return false;
|
|
157
|
+
if (!Number.isInteger(meta.pid) || meta.pid <= 1)
|
|
158
|
+
return true;
|
|
159
|
+
const created = Date.parse(meta.created ?? '');
|
|
160
|
+
if (Number.isFinite(created) && Date.now() - created > MAX_AGE_MS)
|
|
161
|
+
return true;
|
|
162
|
+
return !pidAlive(meta.pid);
|
|
163
|
+
}
|
|
164
|
+
catch {
|
|
165
|
+
// No/unreadable/untrusted marker: a partial copy or a pre-hardening
|
|
166
|
+
// snapshot. Only reap once it is old enough that no live run can
|
|
167
|
+
// plausibly own it.
|
|
168
|
+
try {
|
|
169
|
+
return Date.now() - statSync(path).mtimeMs > LEGACY_STALE_MS;
|
|
170
|
+
}
|
|
171
|
+
catch {
|
|
172
|
+
return false;
|
|
173
|
+
}
|
|
174
|
+
}
|
|
175
|
+
}
|
|
176
|
+
/**
|
|
177
|
+
* Remove snapshots left behind by dead processes (crash, SIGKILL, ENOSPC
|
|
178
|
+
* mid-copy). Sweeps the snapshot base dir, plus os.tmpdir() for snapshots
|
|
179
|
+
* created by pre-hardening versions (UAP_SNAPSHOT_SKIP_TMP_SWEEP=1 disables
|
|
180
|
+
* the legacy sweep). Fail-soft: reaping must never block a delivery run.
|
|
181
|
+
*/
|
|
182
|
+
export function reapStaleSnapshots() {
|
|
183
|
+
const bases = new Set([snapshotBaseDir()]);
|
|
184
|
+
if (process.env.UAP_SNAPSHOT_SKIP_TMP_SWEEP !== '1')
|
|
185
|
+
bases.add(tmpdir());
|
|
186
|
+
for (const base of bases) {
|
|
187
|
+
let entries;
|
|
188
|
+
try {
|
|
189
|
+
entries = readdirSync(base);
|
|
190
|
+
}
|
|
191
|
+
catch {
|
|
192
|
+
continue;
|
|
193
|
+
}
|
|
194
|
+
for (const name of entries) {
|
|
195
|
+
if (!name.startsWith(SNAP_PREFIX))
|
|
196
|
+
continue;
|
|
197
|
+
const path = join(base, name);
|
|
198
|
+
try {
|
|
199
|
+
if (isStaleSnapshot(path))
|
|
200
|
+
rmSync(path, { recursive: true, force: true });
|
|
201
|
+
}
|
|
202
|
+
catch {
|
|
203
|
+
/* fail-soft */
|
|
204
|
+
}
|
|
205
|
+
}
|
|
206
|
+
}
|
|
207
|
+
}
|
|
208
|
+
/**
|
|
209
|
+
* Copy the project tree (minus excluded dirs) to a disk-backed snapshot dir.
|
|
210
|
+
* Returns the snapshot path, or null when the snapshot is skipped (tree over
|
|
211
|
+
* the size cap) or fails for any reason — callers must treat null as "no
|
|
212
|
+
* rollback available for this run", not as an error. Never throws.
|
|
8
213
|
*/
|
|
9
|
-
import { cpSync, rmSync, mkdtempSync, readdirSync } from 'fs';
|
|
10
|
-
import { join, basename } from 'path';
|
|
11
|
-
import { tmpdir } from 'os';
|
|
12
|
-
const EXCLUDE = new Set(['.git', 'node_modules', '.uap-deliver']);
|
|
13
|
-
/** Copy the project tree (minus excluded dirs) to a temp dir; return its path. */
|
|
14
214
|
export function snapshotTree(root) {
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
215
|
+
let snap;
|
|
216
|
+
try {
|
|
217
|
+
reapStaleSnapshots();
|
|
218
|
+
const limit = maxSnapshotBytes();
|
|
219
|
+
if (treeSizeExceeds(root, limit)) {
|
|
220
|
+
console.warn(` no-regress: snapshot skipped — project tree exceeds ${Math.round(limit / (1024 * 1024))} MB ` +
|
|
221
|
+
'(raise UAP_SNAPSHOT_MAX_MB to override)');
|
|
222
|
+
return null;
|
|
223
|
+
}
|
|
224
|
+
snap = mkdtempSync(join(snapshotBaseDir(), SNAP_PREFIX));
|
|
225
|
+
}
|
|
226
|
+
catch (err) {
|
|
227
|
+
console.warn(` no-regress: snapshot unavailable (${err.message}) — rollback disabled for this run`);
|
|
228
|
+
return null;
|
|
229
|
+
}
|
|
230
|
+
try {
|
|
231
|
+
cpSync(root, snap, {
|
|
232
|
+
recursive: true,
|
|
233
|
+
// Never filter the root itself — a project dir named `build`/`dist`/…
|
|
234
|
+
// must still snapshot its contents, not produce an empty snapshot.
|
|
235
|
+
filter: (src) => src === root || !isExcludedDir(src, basename(src)),
|
|
236
|
+
});
|
|
237
|
+
writeFileSync(join(snap, META_FILE), JSON.stringify({ pid: process.pid, created: new Date().toISOString() }));
|
|
238
|
+
return snap;
|
|
239
|
+
}
|
|
240
|
+
catch (err) {
|
|
241
|
+
// Never leak a partial snapshot (the original failure mode: ENOSPC
|
|
242
|
+
// mid-copy left tens of GB behind).
|
|
243
|
+
rmSync(snap, { recursive: true, force: true });
|
|
244
|
+
console.warn(` no-regress: snapshot failed (${err.message}) — rollback disabled for this run`);
|
|
245
|
+
return null;
|
|
246
|
+
}
|
|
247
|
+
}
|
|
248
|
+
/**
|
|
249
|
+
* Delete entries under rootDir that the snapshot doesn't contain (or whose
|
|
250
|
+
* file/dir type flipped), recursing with the same excluded-dir contract as
|
|
251
|
+
* the snapshot filter so nested excluded trees (a workspace's node_modules,
|
|
252
|
+
* a nested target/) survive rollback untouched.
|
|
253
|
+
*/
|
|
254
|
+
function pruneToSnapshot(rootDir, snapDir) {
|
|
255
|
+
for (const entry of readdirSync(rootDir)) {
|
|
256
|
+
const rootPath = join(rootDir, entry);
|
|
257
|
+
let rootStat;
|
|
258
|
+
try {
|
|
259
|
+
rootStat = lstatSync(rootPath);
|
|
260
|
+
}
|
|
261
|
+
catch {
|
|
262
|
+
continue;
|
|
263
|
+
}
|
|
264
|
+
if (rootStat.isDirectory() && EXCLUDE.has(entry))
|
|
26
265
|
continue;
|
|
27
|
-
|
|
266
|
+
let snapStat = null;
|
|
267
|
+
try {
|
|
268
|
+
snapStat = lstatSync(join(snapDir, entry));
|
|
269
|
+
}
|
|
270
|
+
catch {
|
|
271
|
+
/* not in snapshot */
|
|
272
|
+
}
|
|
273
|
+
if (!snapStat) {
|
|
274
|
+
rmSync(rootPath, { recursive: true, force: true });
|
|
275
|
+
}
|
|
276
|
+
else if (rootStat.isDirectory() && snapStat.isDirectory()) {
|
|
277
|
+
pruneToSnapshot(rootPath, join(snapDir, entry));
|
|
278
|
+
}
|
|
279
|
+
else if (rootStat.isDirectory() !== snapStat.isDirectory()) {
|
|
280
|
+
// Type flip (file↔dir): remove so the merge copy can recreate it.
|
|
281
|
+
rmSync(rootPath, { recursive: true, force: true });
|
|
282
|
+
}
|
|
283
|
+
}
|
|
284
|
+
}
|
|
285
|
+
/**
|
|
286
|
+
* Restore the project tree from a snapshot. Validates the snapshot BEFORE
|
|
287
|
+
* touching the project, then prunes extras and merge-copies the snapshot
|
|
288
|
+
* back — copy-first ordering so an interruption leaves a recoverable union
|
|
289
|
+
* of both trees, never a gutted project. On failure the snapshot is marked
|
|
290
|
+
* preserve (reaper-immune) and the error is rethrown.
|
|
291
|
+
*/
|
|
292
|
+
export function restoreTree(root, snap) {
|
|
293
|
+
const entries = readdirSync(snap); // throws if the snapshot is gone — root untouched
|
|
294
|
+
if (!entries.includes(META_FILE)) {
|
|
295
|
+
throw new Error(`refusing to restore from ${snap}: missing ${META_FILE} marker`);
|
|
296
|
+
}
|
|
297
|
+
try {
|
|
298
|
+
pruneToSnapshot(root, snap);
|
|
299
|
+
for (const entry of entries) {
|
|
300
|
+
if (entry === META_FILE)
|
|
301
|
+
continue;
|
|
302
|
+
cpSync(join(snap, entry), join(root, entry), { recursive: true });
|
|
303
|
+
}
|
|
28
304
|
}
|
|
29
|
-
|
|
30
|
-
|
|
305
|
+
catch (err) {
|
|
306
|
+
try {
|
|
307
|
+
writeFileSync(join(snap, META_FILE), JSON.stringify({ pid: process.pid, created: new Date().toISOString(), preserve: true }));
|
|
308
|
+
}
|
|
309
|
+
catch {
|
|
310
|
+
/* marker rewrite is best-effort */
|
|
311
|
+
}
|
|
312
|
+
console.warn(` no-regress: restore FAILED — snapshot preserved at ${snap}`);
|
|
313
|
+
throw err;
|
|
31
314
|
}
|
|
32
315
|
}
|
|
33
316
|
/** Remove a snapshot directory. */
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"snapshot.js","sourceRoot":"","sources":["../../src/delivery/snapshot.ts"],"names":[],"mappings":"AAAA
|
|
1
|
+
{"version":3,"file":"snapshot.js","sourceRoot":"","sources":["../../src/delivery/snapshot.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAEH,OAAO,EACL,MAAM,EACN,MAAM,EACN,WAAW,EACX,SAAS,EACT,WAAW,EACX,YAAY,EACZ,aAAa,EACb,SAAS,EACT,QAAQ,GACT,MAAM,IAAI,CAAC;AACZ,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,OAAO,EAAE,MAAM,MAAM,CAAC;AAC3D,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,IAAI,CAAC;AAErC,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC;IACtB,qBAAqB;IACrB,MAAM;IACN,YAAY;IACZ,cAAc;IACd,cAAc;IACd,KAAK;IACL,cAAc;IACd,MAAM;IACN,OAAO;IACP,KAAK;IACL,UAAU;IACV,OAAO;IACP,OAAO;IACP,aAAa;IACb,QAAQ;IACR,QAAQ;IACR,eAAe;IACf,kBAAkB;IAClB,QAAQ;IACR,SAAS;IACT,QAAQ;IACR,SAAS;IACT,OAAO;IACP,MAAM;IACN,aAAa;IACb,eAAe;IACf,aAAa;IACb,aAAa;IACb,MAAM;CACP,CAAC,CAAC;AAEH,kFAAkF;AAClF,MAAM,SAAS,GAAG,gBAAgB,CAAC;AAEnC,MAAM,WAAW,GAAG,WAAW,CAAC;AAEhC,kFAAkF;AAClF,MAAM,eAAe,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AAE5C,kEAAkE;AAClE,MAAM,UAAU,GAAG,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC;AAE3C,MAAM,cAAc,GAAG,IAAI,CAAC;AAS5B,qFAAqF;AACrF,SAAS,aAAa,CAAC,IAAY,EAAE,IAAY;IAC/C,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACrC,IAAI,CAAC;QACH,OAAO,SAAS,CAAC,IAAI,CAAC,CAAC,WAAW,EAAE,CAAC;IACvC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED,oFAAoF;AACpF,MAAM,UAAU,eAAe;IAC7B,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,gBAAgB,EAAE,IAAI,EAAE,CAAC;IACjD,4EAA4E;IAC5E,mEAAmE;IACnE,MAAM,IAAI,GAAG,GAAG,IAAI,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,WAAW,CAAC,CAAC;IACnG,SAAS,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACrC,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,gBAAgB;IACvB,MAAM,EAAE,GAAG,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;IACnD,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,cAAc,CAAC,GAAG,IAAI,GAAG,IAAI,CAAC;AAC7E,CAAC;AAED;;;;GAIG;AACH,SAAS,eAAe,CAAC,IAAY,EAAE,KAAa;IAClD,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,MAAM,KAAK,GAAG,CAAC,IAAI,CAAC,CAAC;IACrB,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,EAAG,CAAC;QACzB,IAAI,OAAiB,CAAC;QACtB,IAAI,CAAC;YACH,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;QAC7B,CAAC;QAAC,MAAM,CAAC;YACP,SAAS,CAAC,uDAAuD;QACnE,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,OAAO,EAAE,CAAC;YAC3B,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;YAC7B,IAAI,EAAE,CAAC;YACP,IAAI,CAAC;gBACH,EAAE,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;YACvB,CAAC;YAAC,MAAM,CAAC;gBACP,SAAS;YACX,CAAC;YACD,IAAI,EAAE,CAAC,cAAc,EAAE;gBAAE,SAAS;YAClC,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC;gBACrB,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC;oBAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC3C,CAAC;iBAAM,CAAC;gBACN,KAAK,IAAI,EAAE,CAAC,IAAI,CAAC;gBACjB,IAAI,KAAK,GAAG,KAAK;oBAAE,OAAO,IAAI,CAAC;YACjC,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,QAAQ,CAAC,GAAW;IAC3B,IAAI,CAAC;QACH,OAAO,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC;QACrB,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,iDAAiD;QACjD,OAAQ,GAA6B,CAAC,IAAI,KAAK,OAAO,CAAC;IACzD,CAAC;AACH,CAAC;AAED,SAAS,eAAe,CAAC,IAAY;IACnC,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QACvC,qEAAqE;QACrE,iEAAiE;QACjE,MAAM,EAAE,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC;QAC/B,IAAI,CAAC,EAAE,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,IAAI,GAAG,IAAI;YAAE,MAAM,IAAI,KAAK,CAAC,kBAAkB,CAAC,CAAC;QACxE,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAa,CAAC;QACrE,IAAI,IAAI,CAAC,QAAQ,KAAK,IAAI;YAAE,OAAO,KAAK,CAAC;QACzC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,GAAG,CAAC,IAAK,IAAI,CAAC,GAAc,IAAI,CAAC;YAAE,OAAO,IAAI,CAAC;QAC1E,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;QAC/C,IAAI,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,GAAG,UAAU;YAAE,OAAO,IAAI,CAAC;QAC/E,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAa,CAAC,CAAC;IACvC,CAAC;IAAC,MAAM,CAAC;QACP,oEAAoE;QACpE,iEAAiE;QACjE,oBAAoB;QACpB,IAAI,CAAC;YACH,OAAO,IAAI,CAAC,GAAG,EAAE,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,OAAO,GAAG,eAAe,CAAC;QAC/D,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,KAAK,CAAC;QACf,CAAC;IACH,CAAC;AACH,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,kBAAkB;IAChC,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,CAAC,eAAe,EAAE,CAAC,CAAC,CAAC;IAC3C,IAAI,OAAO,CAAC,GAAG,CAAC,2BAA2B,KAAK,GAAG;QAAE,KAAK,CAAC,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC;IACzE,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,OAAiB,CAAC;QACtB,IAAI,CAAC;YACH,OAAO,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC;QAC9B,CAAC;QAAC,MAAM,CAAC;YACP,SAAS;QACX,CAAC;QACD,KAAK,MAAM,IAAI,IAAI,OAAO,EAAE,CAAC;YAC3B,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;gBAAE,SAAS;YAC5C,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YAC9B,IAAI,CAAC;gBACH,IAAI,eAAe,CAAC,IAAI,CAAC;oBAAE,MAAM,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;YAC5E,CAAC;YAAC,MAAM,CAAC;gBACP,eAAe;YACjB,CAAC;QACH,CAAC;IACH,CAAC;AACH,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,YAAY,CAAC,IAAY;IACvC,IAAI,IAAY,CAAC;IACjB,IAAI,CAAC;QACH,kBAAkB,EAAE,CAAC;QACrB,MAAM,KAAK,GAAG,gBAAgB,EAAE,CAAC;QACjC,IAAI,eAAe,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC;YACjC,OAAO,CAAC,IAAI,CACV,yDAAyD,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,MAAM;gBAC9F,yCAAyC,CAC5C,CAAC;YACF,OAAO,IAAI,CAAC;QACd,CAAC;QACD,IAAI,GAAG,WAAW,CAAC,IAAI,CAAC,eAAe,EAAE,EAAE,WAAW,CAAC,CAAC,CAAC;IAC3D,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,OAAO,CAAC,IAAI,CAAC,uCAAwC,GAAa,CAAC,OAAO,oCAAoC,CAAC,CAAC;QAChH,OAAO,IAAI,CAAC;IACd,CAAC;IACD,IAAI,CAAC;QACH,MAAM,CAAC,IAAI,EAAE,IAAI,EAAE;YACjB,SAAS,EAAE,IAAI;YACf,sEAAsE;YACtE,mEAAmE;YACnE,MAAM,EAAE,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,KAAK,IAAI,IAAI,CAAC,aAAa,CAAC,GAAG,EAAE,QAAQ,CAAC,GAAG,CAAC,CAAC;SACpE,CAAC,CAAC;QACH,aAAa,CAAC,IAAI,CAAC,IAAI,EAAE,SAAS,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE,OAAO,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC,CAAC;QAC9G,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,mEAAmE;QACnE,oCAAoC;QACpC,MAAM,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QAC/C,OAAO,CAAC,IAAI,CAAC,kCAAmC,GAAa,CAAC,OAAO,oCAAoC,CAAC,CAAC;QAC3G,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED;;;;;GAKG;AACH,SAAS,eAAe,CAAC,OAAe,EAAE,OAAe;IACvD,KAAK,MAAM,KAAK,IAAI,WAAW,CAAC,OAAO,CAAC,EAAE,CAAC;QACzC,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC;QACtC,IAAI,QAAQ,CAAC;QACb,IAAI,CAAC;YACH,QAAQ,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC;QACjC,CAAC;QAAC,MAAM,CAAC;YACP,SAAS;QACX,CAAC;QACD,IAAI,QAAQ,CAAC,WAAW,EAAE,IAAI,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC;YAAE,SAAS;QAC3D,IAAI,QAAQ,GAAG,IAAI,CAAC;QACpB,IAAI,CAAC;YACH,QAAQ,GAAG,SAAS,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC;QAC7C,CAAC;QAAC,MAAM,CAAC;YACP,qBAAqB;QACvB,CAAC;QACD,IAAI,CAAC,QAAQ,EAAE,CAAC;YACd,MAAM,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACrD,CAAC;aAAM,IAAI,QAAQ,CAAC,WAAW,EAAE,IAAI,QAAQ,CAAC,WAAW,EAAE,EAAE,CAAC;YAC5D,eAAe,CAAC,QAAQ,EAAE,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC;QAClD,CAAC;aAAM,IAAI,QAAQ,CAAC,WAAW,EAAE,KAAK,QAAQ,CAAC,WAAW,EAAE,EAAE,CAAC;YAC7D,kEAAkE;YAClE,MAAM,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACrD,CAAC;IACH,CAAC;AACH,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,WAAW,CAAC,IAAY,EAAE,IAAY;IACpD,MAAM,OAAO,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,kDAAkD;IACrF,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;QACjC,MAAM,IAAI,KAAK,CAAC,4BAA4B,IAAI,aAAa,SAAS,SAAS,CAAC,CAAC;IACnF,CAAC;IACD,IAAI,CAAC;QACH,eAAe,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;QAC5B,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,IAAI,KAAK,KAAK,SAAS;gBAAE,SAAS;YAClC,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,IAAI,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACpE,CAAC;IACH,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,IAAI,CAAC;YACH,aAAa,CACX,IAAI,CAAC,IAAI,EAAE,SAAS,CAAC,EACrB,IAAI,CAAC,SAAS,CAAC,EAAE,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE,OAAO,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CACxF,CAAC;QACJ,CAAC;QAAC,MAAM,CAAC;YACP,mCAAmC;QACrC,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,wDAAwD,IAAI,EAAE,CAAC,CAAC;QAC7E,MAAM,GAAG,CAAC;IACZ,CAAC;AACH,CAAC;AAED,mCAAmC;AACnC,MAAM,UAAU,eAAe,CAAC,IAAY;IAC1C,MAAM,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;AACjD,CAAC"}
|
package/docs/guides/DELIVER.md
CHANGED
|
@@ -137,6 +137,19 @@ uap deliver "add the orders endpoint" --until-deployed
|
|
|
137
137
|
| `--project-root <path>` | Project whose gates define delivery (default: cwd) |
|
|
138
138
|
| `--dry-run` | Show detected gates and plan without calling the model |
|
|
139
139
|
| `--json` | Emit a JSON result |
|
|
140
|
+
| `--keep-best` | Never regress: snapshot the project first, roll back if the run ends with a worse required-gate score |
|
|
141
|
+
|
|
142
|
+
### `--keep-best` snapshots
|
|
143
|
+
|
|
144
|
+
The pre-run snapshot lands on real disk under `~/.cache/uap/snapshots`
|
|
145
|
+
(`UAP_SNAPSHOT_DIR` overrides; absolute paths only) — never `/tmp`, which is
|
|
146
|
+
RAM-backed on many Linux systems. Derived directories (`.git`,
|
|
147
|
+
`node_modules`, `target`, `.venv`, `dist`, `build`, …) are neither
|
|
148
|
+
snapshotted nor touched by a rollback, at any depth. Trees larger than
|
|
149
|
+
`UAP_SNAPSHOT_MAX_MB` (default 4096) skip the snapshot and the run proceeds
|
|
150
|
+
with rollback disabled. Snapshots orphaned by killed runs are reaped
|
|
151
|
+
automatically on the next `--keep-best` run; a snapshot whose restore failed
|
|
152
|
+
is preserved and its path printed.
|
|
140
153
|
|
|
141
154
|
---
|
|
142
155
|
|
|
@@ -89,6 +89,8 @@ Top level: `version`, `project`, `memory`, `worktree`, `costOptimization`,
|
|
|
89
89
|
| `UAP_PARALLEL` | `false` disables parallelism | enabled |
|
|
90
90
|
| `UAP_BENCHMARK_MODE` | `true` enables benchmark template mode | off |
|
|
91
91
|
| `UAP_BENCHMARK_PARALLEL` | Parallel model count in benchmarks | — |
|
|
92
|
+
| `UAP_SNAPSHOT_DIR` | Base dir for `--keep-best` snapshots (absolute path) | `~/.cache/uap/snapshots` |
|
|
93
|
+
| `UAP_SNAPSHOT_MAX_MB` | Skip the `--keep-best` snapshot (rollback disabled) when the tree exceeds this size | `4096` |
|
|
92
94
|
| `HERMES_HOME` | Hermes home dir | `~/.hermes` |
|
|
93
95
|
| `FACTORY_PROJECT_DIR` | Project dir in Factory hook commands | — |
|
|
94
96
|
| `FORGE_UAP_PROJECT` | Project dir in ForgeCode hook scripts | `.` |
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@miller-tech/uap",
|
|
3
|
-
"version": "1.128.
|
|
3
|
+
"version": "1.128.2",
|
|
4
4
|
"description": "Autonomous AI agent memory system with CLAUDE.md protocol enforcement",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "dist/index.js",
|
|
@@ -21,7 +21,7 @@
|
|
|
21
21
|
"start": "node dist/bin/cli.js",
|
|
22
22
|
"test": "vitest",
|
|
23
23
|
"test:ci": "vitest run",
|
|
24
|
-
"test:enforcers": "python3 -m unittest tools.agents.tests.test_enforcement_self_protect tools.agents.tests.test_workdir_scope_enforcer tools.agents.tests.test_path_containment tools.agents.tests.test_path_normalizer_hardened tools.agents.tests.test_sandbox_tool_strip tools.agents.tests.test_proxy_env_loader tools.agents.tests.test_stream_required_tool tools.agents.tests.test_tool_call_wrapper_profiles tools.agents.tests.test_tool_convert_cache",
|
|
24
|
+
"test:enforcers": "python3 -m unittest tools.agents.tests.test_enforcement_self_protect tools.agents.tests.test_schema_diff_gate tools.agents.tests.test_workdir_scope_enforcer tools.agents.tests.test_path_containment tools.agents.tests.test_path_normalizer_hardened tools.agents.tests.test_sandbox_tool_strip tools.agents.tests.test_proxy_env_loader tools.agents.tests.test_stream_required_tool tools.agents.tests.test_tool_call_wrapper_profiles tools.agents.tests.test_tool_convert_cache",
|
|
25
25
|
"test:coverage": "vitest --coverage",
|
|
26
26
|
"bench": "vitest --config vitest.bench.config.ts",
|
|
27
27
|
"lint": "eslint src --ext .ts",
|
|
Binary file
|
|
@@ -86,32 +86,40 @@ def main() -> None:
|
|
|
86
86
|
op, args = parse_cli()
|
|
87
87
|
if op not in EDIT_OPS:
|
|
88
88
|
emit(True, "not a file-edit operation")
|
|
89
|
+
return
|
|
89
90
|
|
|
90
91
|
target = args.get("file_path") or args.get("path") or args.get("target") or ""
|
|
91
92
|
if not target:
|
|
92
93
|
emit(True, "no file path in args")
|
|
94
|
+
return
|
|
93
95
|
|
|
94
96
|
root = repo_root()
|
|
95
97
|
try:
|
|
96
98
|
rel = str(Path(target).resolve().relative_to(root))
|
|
97
99
|
except ValueError:
|
|
98
100
|
emit(True, "target outside repo")
|
|
101
|
+
return
|
|
99
102
|
|
|
100
103
|
rel_posix = rel.replace(os.sep, "/")
|
|
101
104
|
low = rel_posix.lower()
|
|
102
105
|
|
|
103
106
|
if not low.endswith(SOURCE_EXTS):
|
|
104
107
|
emit(True, "not source code")
|
|
108
|
+
return
|
|
105
109
|
if any(rel_posix.startswith(p) for p in EXEMPT_PREFIXES):
|
|
106
110
|
emit(True, f"exempt path: {rel_posix}")
|
|
111
|
+
return
|
|
107
112
|
if any(m in "/" + low for m in TEST_MARKERS):
|
|
108
113
|
emit(True, "test file (protected by deliver itself)")
|
|
114
|
+
return
|
|
109
115
|
|
|
110
116
|
# Escape hatches.
|
|
111
117
|
if os.environ.get("UAP_DELIVER_ACTIVE") == "1":
|
|
112
118
|
emit(True, "inside a deliver-driven run")
|
|
119
|
+
return
|
|
113
120
|
if os.environ.get("UAP_DELIVER_BYPASS") == "1":
|
|
114
121
|
emit(True, "UAP_DELIVER_BYPASS override set")
|
|
122
|
+
return
|
|
115
123
|
|
|
116
124
|
# #3-F: terse, imperative, model-parseable. Weak local models otherwise
|
|
117
125
|
# retry the blocked edit or hallucinate completion ("the files exist") when
|
|
@@ -143,6 +151,7 @@ def main() -> None:
|
|
|
143
151
|
route="deliver",
|
|
144
152
|
deliverHint=f'implement the intended change to {rel_posix}',
|
|
145
153
|
)
|
|
154
|
+
return
|
|
146
155
|
|
|
147
156
|
# Advisory (opt-out): never blocks. Surface the nudge, then allow.
|
|
148
157
|
print(f"[delivery-enforcement advisory] {msg}", file=sys.stderr)
|
|
@@ -150,4 +159,4 @@ def main() -> None:
|
|
|
150
159
|
|
|
151
160
|
|
|
152
161
|
if __name__ == "__main__":
|
|
153
|
-
main()
|
|
162
|
+
main()
|
|
@@ -16,7 +16,10 @@ WATCHED_RE = re.compile(
|
|
|
16
16
|
r"infra/helm_charts/[^/]*envoy|infra/helm_charts/[^/]*sentinel)",
|
|
17
17
|
re.I,
|
|
18
18
|
)
|
|
19
|
-
|
|
19
|
+
# NOTE: bare "Bash" used to be in this set, which made EVERY shell command a
|
|
20
|
+
# gate point — including the `uap schema-diff` remedy itself (self-deadlock).
|
|
21
|
+
# Gate only actual commit/push commands (main() also inspects cmd content).
|
|
22
|
+
COMMIT_OPS = {"git-commit", "git commit"}
|
|
20
23
|
RECENT_SEC = 3600
|
|
21
24
|
|
|
22
25
|
|
|
@@ -35,17 +38,31 @@ def schema_diff_ok(root: Path) -> bool:
|
|
|
35
38
|
return False
|
|
36
39
|
try:
|
|
37
40
|
con = sqlite3.connect(f"file:{db}?mode=ro", uri=True, timeout=1.0)
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
)
|
|
43
|
-
|
|
41
|
+
# `uap memory store` writes to `memories` (type 'action'), while older
|
|
42
|
+
# UAP wrote session rows to `session_memories` — accept the marker from
|
|
43
|
+
# either table so the documented remedy actually clears the gate.
|
|
44
|
+
row = None
|
|
45
|
+
for table in ("memories", "session_memories"):
|
|
46
|
+
try:
|
|
47
|
+
cur = con.execute(
|
|
48
|
+
f"SELECT timestamp FROM {table} "
|
|
49
|
+
"WHERE content LIKE '%schema-diff%pass%' "
|
|
50
|
+
"ORDER BY id DESC LIMIT 1"
|
|
51
|
+
)
|
|
52
|
+
r = cur.fetchone()
|
|
53
|
+
if r and (row is None or r[0] > row[0]):
|
|
54
|
+
row = r
|
|
55
|
+
except sqlite3.Error:
|
|
56
|
+
continue
|
|
44
57
|
con.close()
|
|
45
58
|
if not row:
|
|
46
59
|
return False
|
|
47
60
|
try:
|
|
48
|
-
|
|
61
|
+
# Timestamps are stored as UTC ISO-8601 (trailing 'Z'); parse them
|
|
62
|
+
# as UTC — time.mktime would misread them as local time and expire
|
|
63
|
+
# the marker hours early (or late) depending on the host TZ.
|
|
64
|
+
import calendar
|
|
65
|
+
ts = calendar.timegm(time.strptime(row[0][:19], "%Y-%m-%dT%H:%M:%S"))
|
|
49
66
|
except Exception: # noqa: BLE001
|
|
50
67
|
return False
|
|
51
68
|
return (time.time() - ts) < RECENT_SEC
|
|
Binary file
|
|
@@ -0,0 +1,115 @@
|
|
|
1
|
+
"""Tests for the schema-diff-gate enforcer fixes (self-deadlock + marker read).
|
|
2
|
+
|
|
3
|
+
Covers the 2026-07-08 mid-session incident: bare "Bash" in COMMIT_OPS gated
|
|
4
|
+
EVERY shell command (including the `uap schema-diff` remedy — self-deadlock),
|
|
5
|
+
and the pass-marker check read only the legacy session_memories table and
|
|
6
|
+
parsed UTC timestamps with time.mktime (host-timezone skew).
|
|
7
|
+
"""
|
|
8
|
+
|
|
9
|
+
from __future__ import annotations
|
|
10
|
+
|
|
11
|
+
import json
|
|
12
|
+
import os
|
|
13
|
+
import sqlite3
|
|
14
|
+
import subprocess
|
|
15
|
+
import tempfile
|
|
16
|
+
import unittest
|
|
17
|
+
from datetime import datetime, timedelta, timezone
|
|
18
|
+
from pathlib import Path
|
|
19
|
+
|
|
20
|
+
REPO = Path(__file__).resolve().parents[3]
|
|
21
|
+
ENFORCER = REPO / "src" / "policies" / "enforcers" / "schema_diff_gate.py"
|
|
22
|
+
|
|
23
|
+
|
|
24
|
+
def run_gate(op: str, args: dict, root: Path, tz: str | None = None):
|
|
25
|
+
env = {k: v for k, v in os.environ.items() if not k.startswith("GIT_")}
|
|
26
|
+
env["UAP_REPO_ROOT"] = str(root)
|
|
27
|
+
env["UAP_WORKTREE_ROOT"] = str(root)
|
|
28
|
+
if tz:
|
|
29
|
+
env["TZ"] = tz
|
|
30
|
+
r = subprocess.run(
|
|
31
|
+
["python3", str(ENFORCER), "--operation", op, "--args", json.dumps(args)],
|
|
32
|
+
cwd=root,
|
|
33
|
+
env=env,
|
|
34
|
+
capture_output=True,
|
|
35
|
+
text=True,
|
|
36
|
+
timeout=30,
|
|
37
|
+
)
|
|
38
|
+
try:
|
|
39
|
+
payload = json.loads(r.stdout or "{}")
|
|
40
|
+
except json.JSONDecodeError:
|
|
41
|
+
payload = {}
|
|
42
|
+
return r.returncode, payload.get("allowed", False), payload.get("reason", "")
|
|
43
|
+
|
|
44
|
+
|
|
45
|
+
class SchemaDiffGateTest(unittest.TestCase):
|
|
46
|
+
def setUp(self):
|
|
47
|
+
self._tmp = tempfile.TemporaryDirectory(prefix="schema-gate-")
|
|
48
|
+
self.root = Path(self._tmp.name)
|
|
49
|
+
subprocess.run(["git", "init", "-q"], cwd=self.root, check=True)
|
|
50
|
+
subprocess.run(
|
|
51
|
+
["git", "-c", "user.email=t@t", "-c", "user.name=t", "commit", "-q", "--allow-empty", "-m", "init"],
|
|
52
|
+
cwd=self.root,
|
|
53
|
+
check=True,
|
|
54
|
+
)
|
|
55
|
+
(self.root / "migrations").mkdir()
|
|
56
|
+
(self.root / "migrations" / "001_add_table.sql").write_text("CREATE TABLE t (id int);")
|
|
57
|
+
subprocess.run(["git", "add", "migrations"], cwd=self.root, check=True)
|
|
58
|
+
|
|
59
|
+
def tearDown(self):
|
|
60
|
+
self._tmp.cleanup()
|
|
61
|
+
|
|
62
|
+
def write_marker(self, iso_timestamp: str, table: str = "memories"):
|
|
63
|
+
mem = self.root / "agents" / "data" / "memory"
|
|
64
|
+
mem.mkdir(parents=True)
|
|
65
|
+
con = sqlite3.connect(mem / "short_term.db")
|
|
66
|
+
con.execute(f"CREATE TABLE {table} (id INTEGER PRIMARY KEY, content TEXT, timestamp TEXT)")
|
|
67
|
+
con.execute(
|
|
68
|
+
f"INSERT INTO {table} (content, timestamp) VALUES (?, ?)",
|
|
69
|
+
("schema-diff pass: verified", iso_timestamp),
|
|
70
|
+
)
|
|
71
|
+
con.commit()
|
|
72
|
+
con.close()
|
|
73
|
+
|
|
74
|
+
def test_ordinary_bash_is_not_a_gate_point(self):
|
|
75
|
+
"""Regression: bare Bash op must not gate — that deadlocked the remedy itself."""
|
|
76
|
+
code, allowed, reason = run_gate("Bash", {"command": "uap schema-diff"}, self.root)
|
|
77
|
+
self.assertEqual(code, 0)
|
|
78
|
+
self.assertTrue(allowed)
|
|
79
|
+
self.assertIn("not a commit/push gate point", reason)
|
|
80
|
+
|
|
81
|
+
def test_commit_command_still_gated(self):
|
|
82
|
+
code, allowed, reason = run_gate("Bash", {"command": 'git commit -m "add migration"'}, self.root)
|
|
83
|
+
self.assertEqual(code, 2)
|
|
84
|
+
self.assertFalse(allowed)
|
|
85
|
+
self.assertIn("uap schema-diff", reason)
|
|
86
|
+
|
|
87
|
+
def test_fresh_marker_in_memories_table_clears_gate(self):
|
|
88
|
+
"""`uap memory store` writes to `memories`; the gate must accept it."""
|
|
89
|
+
self.write_marker(datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"))
|
|
90
|
+
code, allowed, reason = run_gate("git-commit", {}, self.root)
|
|
91
|
+
self.assertEqual(code, 0)
|
|
92
|
+
self.assertTrue(allowed)
|
|
93
|
+
self.assertIn("recent schema-diff pass", reason)
|
|
94
|
+
|
|
95
|
+
def test_legacy_session_memories_table_still_accepted(self):
|
|
96
|
+
self.write_marker(
|
|
97
|
+
datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"), table="session_memories"
|
|
98
|
+
)
|
|
99
|
+
code, allowed, _ = run_gate("git-commit", {}, self.root)
|
|
100
|
+
self.assertEqual(code, 0)
|
|
101
|
+
self.assertTrue(allowed)
|
|
102
|
+
|
|
103
|
+
def test_stale_marker_blocks_regardless_of_host_timezone(self):
|
|
104
|
+
"""Pre-fix, time.mktime read UTC timestamps as local: east-of-UTC hosts
|
|
105
|
+
saw stale markers shifted into the future and the gate cleared on a
|
|
106
|
+
2h-old pass."""
|
|
107
|
+
stale = (datetime.now(timezone.utc) - timedelta(hours=2)).strftime("%Y-%m-%dT%H:%M:%SZ")
|
|
108
|
+
self.write_marker(stale)
|
|
109
|
+
code, allowed, _ = run_gate("git-commit", {}, self.root, tz="Australia/Sydney")
|
|
110
|
+
self.assertEqual(code, 2)
|
|
111
|
+
self.assertFalse(allowed)
|
|
112
|
+
|
|
113
|
+
|
|
114
|
+
if __name__ == "__main__":
|
|
115
|
+
unittest.main()
|