qualiow-exploratory-testing 2.3.0 → 2.4.0
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/.claude-plugin/marketplace.json +1 -1
- package/.claude-plugin/plugin.json +1 -1
- package/CHANGELOG.md +177 -0
- package/README.md +71 -21
- package/agents/qa-bug-judge.md +1 -1
- package/agents/qa-diff-indexer-agent.md +1 -1
- package/agents/qa-gather-agent.md +1 -1
- package/agents/qa-page-mapper-agent.md +3 -1
- package/agents/qa-reporting-agent.md +28 -14
- package/bin/mobile-cli.mjs +74 -26
- package/bin/qualiow +7 -2
- package/bin/wk-ios +60 -6
- package/bin/wkeval.mjs +7 -2
- package/data/security/SECURITY-POLICY.md +22 -3
- package/data/targets/parabank.yml +13 -0
- package/data/targets/saucedemo.yml +13 -0
- package/data/templates/bug-report.md +3 -0
- package/data/templates/session-charter.md +7 -5
- package/data/templates/session-report.md +7 -1
- package/dist/cli/index.js +2060 -361
- package/dist/index.d.ts +813 -26
- package/dist/index.js +2694 -704
- package/docs/GETTING-STARTED.md +71 -17
- package/docs/MOBILE-SETUP.md +31 -0
- package/hooks/hooks.json +12 -2
- package/hooks/scripts/bash-guard.mjs +242 -0
- package/hooks/scripts/secret-patterns.d.mts +4 -1
- package/hooks/scripts/secret-patterns.mjs +126 -34
- package/hooks/scripts/write-guard.mjs +70 -15
- package/package.json +3 -2
- package/skills/qa-explore/SKILL.md +10 -3
- package/skills/qa-explore/phases/00-setup.md +40 -3
- package/skills/qa-explore/phases/02-charter.md +25 -5
- package/skills/qa-explore/phases/04-journeys.md +14 -0
- package/skills/qa-explore/phases/05-features.md +10 -0
- package/skills/qa-explore/phases/06-edge-cases.md +10 -0
- package/skills/qa-explore/phases/07-reporting.md +93 -11
- package/skills/qa-explore/references/delegation-rules.md +12 -1
- package/skills/qa-explore/references/evidence-triage.md +44 -12
- package/skills/qa-explore/references/output-contract.md +69 -7
- package/skills/qa-explore/references/security-rules.md +25 -3
- package/skills/qa-explore-cleanup/SKILL.md +1 -1
- package/skills/qa-explore-mobile/SKILL.md +13 -3
- package/skills/qa-explore-mobile/phases/02-charter.md +2 -1
- package/skills/qa-explore-mobile/phases/07-reporting.md +2 -0
- package/skills/qa-explore-mobile/references/mobile-edges.md +3 -1
- package/skills/qa-explore-quick/SKILL.md +48 -9
- package/skills/qa-explore-report/SKILL.md +6 -2
- package/skills/qa-gather/SKILL.md +1 -1
- package/skills/qa-knowledge-add/SKILL.md +1 -1
- package/skills/qa-knowledge-list/SKILL.md +1 -1
- package/skills/qa-verify-backend/SKILL.md +1 -1
- package/skills/qa-verify-backend/phases/05-reporting.md +2 -1
- package/data/targets/testers-ai.yml +0 -14
package/docs/GETTING-STARTED.md
CHANGED
|
@@ -32,7 +32,7 @@ That writes:
|
|
|
32
32
|
| the 11 skills | `.claude/skills/qa-*/` |
|
|
33
33
|
| the five sub-agents | `.claude/agents/` |
|
|
34
34
|
| knowledge base, domain profiles, templates, security policy | `data/knowledge/`, `data/domains/`, `data/templates/`, `data/security/` |
|
|
35
|
-
| the `_default` target (add `--include-examples` for the `_example-*` templates and `
|
|
35
|
+
| the `_default` target (add `--include-examples` for the `_example-*` templates and the `parabank.yml` and `saucedemo.yml` demo targets) | `data/targets/` |
|
|
36
36
|
| the mobile driver plus `setup-mobile.sh` and `doctor-mobile.sh`, executable | `qa/bin/` |
|
|
37
37
|
| the credential template | `qa/.env.example` |
|
|
38
38
|
| empty session/bug/context trees and the auth directory | `output/sessions/`, `output/bugs/`, `output/context/`, `.auth/` |
|
|
@@ -52,18 +52,22 @@ npx playwright-cli install --skills # optional: the official Playwright skill
|
|
|
52
52
|
qualiow init --hooks
|
|
53
53
|
```
|
|
54
54
|
|
|
55
|
-
Optional, and only for npm projects — a plugin install already ships the same
|
|
55
|
+
Optional, and only for npm projects — a plugin install already ships the same three hooks and
|
|
56
56
|
has them on by default, so do not enable both in one project. `--hooks` copies the guard
|
|
57
|
-
scripts to `qa/hooks/` and merges into `.claude/settings.json`
|
|
58
|
-
three `permissions.allow` rules.
|
|
59
|
-
|
|
57
|
+
scripts to `qa/hooks/` and merges into `.claude/settings.json` three `PreToolUse` entries and
|
|
58
|
+
three `permissions.allow` rules. Running it again changes nothing, and hooks you already had
|
|
59
|
+
are kept; an older qualiow-only write-guard entry gets the new matcher. If `settings.json`
|
|
60
|
+
does not parse (a comment, a trailing comma), init stops without writing anything and prints
|
|
61
|
+
the entries to merge by hand.
|
|
60
62
|
|
|
61
63
|
What the guards do: `read-guard.mjs` denies a whole-file `Read` of a qualiow file over
|
|
62
64
|
`QUALIOW_READ_MAX_LINES` lines (default 300) — the knowledge manifest, a knowledge release
|
|
63
65
|
file, a session phase file or a raw snapshot — and names the cheaper route in the refusal; a
|
|
64
66
|
`Read` with `offset` or `limit` passes. `write-guard.mjs` refuses a `Write` or `Edit` under
|
|
65
|
-
`output/` whose content matches the redaction list, so a secret never reaches disk.
|
|
66
|
-
|
|
67
|
+
`output/` whose content matches the redaction list, so a secret never reaches disk.
|
|
68
|
+
`bash-guard.mjs` sees every `Bash` command but only acts on a few: it asks before
|
|
69
|
+
`playwright-cli run-code` and before `git` options that can run a shell, and denies `npx -c`
|
|
70
|
+
or `npx -y` of an unknown package. `QUALIOW_HOOKS=off` disables all three.
|
|
67
71
|
|
|
68
72
|
To write it by hand instead, this is the whole of it:
|
|
69
73
|
|
|
@@ -78,10 +82,16 @@ To write it by hand instead, this is the whole of it:
|
|
|
78
82
|
]
|
|
79
83
|
},
|
|
80
84
|
{
|
|
81
|
-
"matcher": "Write|Edit|MultiEdit",
|
|
85
|
+
"matcher": "Write|Edit|MultiEdit|NotebookEdit",
|
|
82
86
|
"hooks": [
|
|
83
87
|
{ "type": "command", "command": "node \"$CLAUDE_PROJECT_DIR/qa/hooks/write-guard.mjs\"" }
|
|
84
88
|
]
|
|
89
|
+
},
|
|
90
|
+
{
|
|
91
|
+
"matcher": "Bash",
|
|
92
|
+
"hooks": [
|
|
93
|
+
{ "type": "command", "command": "node \"$CLAUDE_PROJECT_DIR/qa/hooks/bash-guard.mjs\"" }
|
|
94
|
+
]
|
|
85
95
|
}
|
|
86
96
|
]
|
|
87
97
|
},
|
|
@@ -123,11 +133,13 @@ of that path needs network. A plain git checkout is described in the project REA
|
|
|
123
133
|
|
|
124
134
|
### Option A: explore a public site
|
|
125
135
|
|
|
126
|
-
Point `/qa-explore` at any public site.
|
|
127
|
-
|
|
136
|
+
Point `/qa-explore` at any public site. Two public demo apps ship as target configs (install
|
|
137
|
+
them with `npx qualiow init --include-examples`): `parabank`, a demo online bank
|
|
138
|
+
(<https://parabank.parasoft.com/>), and `saucedemo`, a small e-commerce shop
|
|
139
|
+
(<https://www.saucedemo.com/>).
|
|
128
140
|
|
|
129
141
|
```bash
|
|
130
|
-
/qa-explore
|
|
142
|
+
/qa-explore --target parabank
|
|
131
143
|
```
|
|
132
144
|
|
|
133
145
|
Claude will:
|
|
@@ -141,8 +153,10 @@ Claude will:
|
|
|
141
153
|
6. Draft one bug report per finding, then have the `qa-bug-judge` sub-agent try to refute each
|
|
142
154
|
one from a fresh context; survivors go to `bugs/`, refuted candidates to `bugs/refuted/`
|
|
143
155
|
7. Write `phase-7-notes.md` — the executive summary, coverage map, recommendations and
|
|
144
|
-
reflection —
|
|
145
|
-
`qualiow session
|
|
156
|
+
reflection — and `stats.json` with one `coverage.areas` entry per charter area, run
|
|
157
|
+
`qualiow session level <session-dir> --write` to compute the coverage level, then hand the
|
|
158
|
+
assembly of `session-report.md` and the `qualiow session finalize <session-dir>` call to
|
|
159
|
+
the `qa-reporting-agent` sub-agent
|
|
146
160
|
|
|
147
161
|
The eight phases and their share of the 45-minute cap:
|
|
148
162
|
|
|
@@ -174,7 +188,7 @@ phase 7; a bug it cannot reach (model unavailable, timeout, budget spent) still
|
|
|
174
188
|
`Unverified`. To skip it for one run:
|
|
175
189
|
|
|
176
190
|
```bash
|
|
177
|
-
/qa-explore
|
|
191
|
+
/qa-explore --target saucedemo --no-judge
|
|
178
192
|
```
|
|
179
193
|
|
|
180
194
|
or set it off for a target in its YAML:
|
|
@@ -186,6 +200,20 @@ verification:
|
|
|
186
200
|
|
|
187
201
|
With the judge off, phase 7 writes `bugs/` directly and no `verification/` directory appears.
|
|
188
202
|
|
|
203
|
+
The 45-minute cap means some areas are not reached. They are not dropped: the session marks
|
|
204
|
+
them `deferred` in `stats.json`, the CLI renders them into `backlog.md`, and the next session
|
|
205
|
+
can start from them:
|
|
206
|
+
|
|
207
|
+
```bash
|
|
208
|
+
/qa-explore --target my-app --continue latest
|
|
209
|
+
```
|
|
210
|
+
|
|
211
|
+
`--continue` takes a session directory name or `latest`, and only a finalized session of the
|
|
212
|
+
same target. The new session never opens the old one's files; it runs
|
|
213
|
+
`qualiow session continue-check latest --target my-app`, which prints a fenced, redacted,
|
|
214
|
+
path-only summary (risk rows, backlog, site-map paths) that is treated as data, and puts the
|
|
215
|
+
backlog rows first in its charter with their IDs.
|
|
216
|
+
|
|
189
217
|
### Option B: quick check on one page
|
|
190
218
|
|
|
191
219
|
```bash
|
|
@@ -234,9 +262,14 @@ output/sessions/2026-09-08-1813-explore-parabank/
|
|
|
234
262
|
verification/ # claim cards and verdicts (judge on)
|
|
235
263
|
screenshots/
|
|
236
264
|
BUG-001.png
|
|
265
|
+
A1-login.png # per-area evidence, named after the charter ID
|
|
237
266
|
snapshots/ # raw page snapshots — working files, never part of the report
|
|
238
267
|
videos/
|
|
268
|
+
traces/ # Playwright traces kept as evidence
|
|
269
|
+
evidence/ # any other evidence file an area or bug cites
|
|
239
270
|
session-report.md # the deliverable
|
|
271
|
+
evidence-level.md # coverage level, written by `qualiow session level --write`
|
|
272
|
+
backlog.md # areas not tested, rendered by the same command
|
|
240
273
|
stats.json # machine-readable session metrics
|
|
241
274
|
```
|
|
242
275
|
|
|
@@ -259,13 +292,31 @@ qualiow session finalize latest # validate, then append the rows
|
|
|
259
292
|
|
|
260
293
|
Finalizing twice changes nothing — the rows are appended once.
|
|
261
294
|
|
|
295
|
+
Explore and quick sessions from 2.4.0 are **contract 2** (`"contract": 2` in `stats.json`), and
|
|
296
|
+
finalize checks their coverage record too: a `coverage.areas` entry for every charter row,
|
|
297
|
+
evidence files that exist under `screenshots/`, `videos/`, `traces/`, `logs/` or `evidence/`
|
|
298
|
+
(an `A<N>-` screenshot for every tested area), a reason for every area not tested, an
|
|
299
|
+
`**Area:**` line on every bug, and a coverage level that matches all of that. The level is
|
|
300
|
+
computed, never typed:
|
|
301
|
+
|
|
302
|
+
```bash
|
|
303
|
+
qualiow session level latest # validate and print the level
|
|
304
|
+
qualiow session level latest --write # also write evidence-level.md, backlog.md, stats.coverage_level
|
|
305
|
+
```
|
|
306
|
+
|
|
307
|
+
The level — `unassessed`, `incomplete`, `qualified` or `complete`, overall and per risk tier — says
|
|
308
|
+
how much of the charter was covered with evidence. It is a coverage fact, never a ship
|
|
309
|
+
probability or a release verdict; the findings line beside it is separate. Edit the stats, the
|
|
310
|
+
charter or a bug after `--write` and finalize asks you to run it again. Sessions written before
|
|
311
|
+
2.4.0, and mobile and backend sessions, are contract 1 and are checked as before.
|
|
312
|
+
|
|
262
313
|
### Bug report structure
|
|
263
314
|
|
|
264
315
|
Each report opens with the confidentiality blockquote, then:
|
|
265
316
|
|
|
266
317
|
- **Title** — `# BUG-NNN: [Component] fails [Condition] causing [Impact]`
|
|
267
318
|
- **Severity** — Critical, High, Medium or Low (conservative by design) — plus Priority,
|
|
268
|
-
Component, URL, Environment and reproduction rate
|
|
319
|
+
Component, Area (the charter ID, explore and quick), URL, Environment and reproduction rate
|
|
269
320
|
- **Summary**, **Expected Behavior**, **Actual Behavior**
|
|
270
321
|
- **Steps to Reproduce** — numbered, specific enough for someone else to follow
|
|
271
322
|
- **Business Impact** — revenue, trust, regulatory, data and scale. Mandatory
|
|
@@ -275,6 +326,7 @@ Each report opens with the confidentiality blockquote, then:
|
|
|
275
326
|
### Session report structure
|
|
276
327
|
|
|
277
328
|
- Session metadata and executive summary
|
|
329
|
+
- **Coverage level** (contract 2) — `evidence-level.md` copied verbatim
|
|
278
330
|
- Summary stats and the bug table
|
|
279
331
|
- **Coverage map** (`| Area | Risk | Status | Bugs | Notes |`) — including what was *not* tested
|
|
280
332
|
- Observations that are not bugs but are worth discussing
|
|
@@ -406,14 +458,16 @@ decide which entries a skill always gets without touching the skill itself.
|
|
|
406
458
|
|
|
407
459
|
```bash
|
|
408
460
|
qualiow init --dry-run # preview what init would write
|
|
409
|
-
qualiow init --hooks # add the read/write guards to .claude/settings.json
|
|
461
|
+
qualiow init --hooks # add the read/write/bash guards to .claude/settings.json
|
|
410
462
|
qualiow validate --all # targets + qa/target.yml + domains + knowledge base
|
|
411
463
|
qualiow list sessions # also: knowledge | targets | domains
|
|
412
464
|
qualiow list knowledge --stats # also: --domain --tag --type --entry <id> --changelog
|
|
413
|
-
qualiow report -s latest -f html -o report.html
|
|
465
|
+
qualiow report -s latest -f html -o report.html # writes output/report.html
|
|
414
466
|
qualiow kb check
|
|
415
467
|
qualiow kb digest --for explore # the knowledge a session loads
|
|
468
|
+
qualiow session level latest --write # compute a contract-2 session's coverage level
|
|
416
469
|
qualiow session finalize latest # validate a finished session and index it
|
|
470
|
+
qualiow session continue-check latest --target my-app # what --continue would carry forward
|
|
417
471
|
qualiow session prune --older-than 30 # dry-run; add --yes to remove
|
|
418
472
|
```
|
|
419
473
|
|
package/docs/MOBILE-SETUP.md
CHANGED
|
@@ -181,6 +181,37 @@ Requires `ios-webkit-debug-proxy`, `python3`, Node ≥ 22.4 (`wkeval.mjs` uses t
|
|
|
181
181
|
`WebSocket`), and a booted Simulator with Safari on a page. `WK_IOS_PORT` (default 9221) sets
|
|
182
182
|
the iwdp device-list port; pages are served on `WK_IOS_PORT + 1`.
|
|
183
183
|
|
|
184
|
+
> **⚠️ The proxy listens on every network interface, without authentication.**
|
|
185
|
+
> `ios-webkit-debug-proxy` (1.9.x) has no bind-address option, so ports `WK_IOS_PORT` to
|
|
186
|
+
> `WK_IOS_PORT + 101` are reachable from your LAN while it runs, and anyone who can reach them
|
|
187
|
+
> can evaluate JavaScript in the Simulator's pages — including their logged-in sessions. The
|
|
188
|
+
> proxy stays up between calls so each eval is fast; `wk-ios` prints a warning when it starts
|
|
189
|
+
> it. **Run `bin/wk-ios --stop` as soon as the session ends**, keep the macOS firewall on
|
|
190
|
+
> (System Settings → Network → Firewall), and avoid using the bridge on untrusted networks.
|
|
191
|
+
|
|
192
|
+
Safari keeps old tabs inspectable, so the newest page is not always the one under test. Set
|
|
193
|
+
`WK_PAGE_FILTER` to pin it. The filter is matched against the parsed page URL — never as a raw
|
|
194
|
+
substring, so a page cannot win by carrying the filter in its query string or fragment — and the
|
|
195
|
+
newest matching page is used:
|
|
196
|
+
|
|
197
|
+
| `WK_PAGE_FILTER` | Matches |
|
|
198
|
+
|---|---|
|
|
199
|
+
| `/account` | any page whose path starts with `/account` (any host) |
|
|
200
|
+
| `staging.example.com` | that host or any subdomain of it; `host:port` matches that exact host and port |
|
|
201
|
+
| `staging.example.com/account` | exactly that host, path starting with `/account` |
|
|
202
|
+
| `https://staging.example.com/account` | same scheme and host[:port], path starting with `/account` |
|
|
203
|
+
|
|
204
|
+
When no page matches, `wk-ios` stops with an error rather than evaluating in another tab. Prefer
|
|
205
|
+
a host-qualified form: a bare `/path` does not pin the host.
|
|
206
|
+
|
|
207
|
+
```bash
|
|
208
|
+
WK_PAGE_FILTER=staging.example.com/account bin/wk-ios 'document.title'
|
|
209
|
+
```
|
|
210
|
+
|
|
211
|
+
The proxy's pid and log files live in a private per-user directory,
|
|
212
|
+
`${TMPDIR:-/tmp}/wk-ios-<uid>/` (mode 700). Page values printed by `wk-ios`, and app labels
|
|
213
|
+
printed by `mcli snapshot` and `mcli logs`, have terminal control characters stripped.
|
|
214
|
+
|
|
184
215
|
## Why a Google **Play** (or `google_apis`) Android image
|
|
185
216
|
|
|
186
217
|
A stock **AOSP** system image ships **no `com.android.chrome`** — only a basic AOSP
|
package/hooks/hooks.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
{
|
|
2
|
-
"description": "qualiow routing guards: whole-file reads of qualiow data over the line threshold,
|
|
2
|
+
"description": "qualiow routing guards: whole-file reads of qualiow data over the line threshold, secrets in session output, and code-execution routes in Bash (playwright-cli run-code, arbitrary npx)",
|
|
3
3
|
"hooks": {
|
|
4
4
|
"PreToolUse": [
|
|
5
5
|
{
|
|
@@ -13,7 +13,7 @@
|
|
|
13
13
|
]
|
|
14
14
|
},
|
|
15
15
|
{
|
|
16
|
-
"matcher": "Write|Edit|MultiEdit",
|
|
16
|
+
"matcher": "Write|Edit|MultiEdit|NotebookEdit",
|
|
17
17
|
"hooks": [
|
|
18
18
|
{
|
|
19
19
|
"type": "command",
|
|
@@ -21,6 +21,16 @@
|
|
|
21
21
|
"timeout": 10
|
|
22
22
|
}
|
|
23
23
|
]
|
|
24
|
+
},
|
|
25
|
+
{
|
|
26
|
+
"matcher": "Bash",
|
|
27
|
+
"hooks": [
|
|
28
|
+
{
|
|
29
|
+
"type": "command",
|
|
30
|
+
"command": "node \"${CLAUDE_PLUGIN_ROOT}/hooks/scripts/bash-guard.mjs\"",
|
|
31
|
+
"timeout": 10
|
|
32
|
+
}
|
|
33
|
+
]
|
|
24
34
|
}
|
|
25
35
|
]
|
|
26
36
|
}
|
|
@@ -0,0 +1,242 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
/**
|
|
3
|
+
* PreToolUse hook (matcher: Bash).
|
|
4
|
+
*
|
|
5
|
+
* Code-execution routes that a pre-approved `Bash(playwright-cli:*)`,
|
|
6
|
+
* `Bash(git -C:*)` or a permissive `npx` rule would otherwise let through
|
|
7
|
+
* without a prompt:
|
|
8
|
+
*
|
|
9
|
+
* - `playwright-cli run-code` (direct or via `npx playwright-cli`) runs its
|
|
10
|
+
* JavaScript in the Playwright node process, not in the browser sandbox —
|
|
11
|
+
* the `vm` context holding `page` is escapable to the host `process`. It is
|
|
12
|
+
* used legitimately, so the decision is `ask`: the user confirms every call.
|
|
13
|
+
* - `npx -c/--call <shell>`, and `npx -y`/`-p` of a package outside a short
|
|
14
|
+
* allowlist, install or run arbitrary code. The decision is `deny`.
|
|
15
|
+
* - `git -c <key>=<value>`, `--config-env`, `--upload-pack`, `--receive-pack`,
|
|
16
|
+
* `--exec`, `--ext-diff` and `--output` run a program or write an arbitrary
|
|
17
|
+
* file from an otherwise read-only git call — the skills pre-approve
|
|
18
|
+
* `git -C <repo> …` for the repository under test. The decision is `ask`.
|
|
19
|
+
*
|
|
20
|
+
* Compound commands are split on `&&`, `||`, `;`, `|`, `&` and newlines;
|
|
21
|
+
* leading `VAR=value` assignments and `env`/`command`/`exec`/`time`/`nohup`
|
|
22
|
+
* wrappers are skipped; `bash|sh|zsh -c '<cmd>'`, `eval`, `$(…)` and backticks
|
|
23
|
+
* are inspected as nested commands. A `deny` anywhere wins over an `ask`.
|
|
24
|
+
*
|
|
25
|
+
* Node built-ins only — a git-sourced plugin install has neither `dist/` nor
|
|
26
|
+
* `node_modules`. Never throws and never blocks on its own failure: any error
|
|
27
|
+
* here lets the command through (exit 0, no stdout) to the normal permission
|
|
28
|
+
* check rather than breaking the caller because of a hook bug.
|
|
29
|
+
*/
|
|
30
|
+
import { readFileSync } from 'node:fs';
|
|
31
|
+
|
|
32
|
+
const NPX_ALLOWED_PACKAGES = new Set(['playwright-cli', '@playwright/cli', 'qualiow-exploratory-testing']);
|
|
33
|
+
const PLAYWRIGHT_CLI_NAMES = new Set(['playwright-cli', '@playwright/cli']);
|
|
34
|
+
const WRAPPERS = new Set(['env', 'command', 'exec', 'time', 'nohup']);
|
|
35
|
+
const SHELLS = new Set(['bash', 'sh', 'zsh', 'dash']);
|
|
36
|
+
const MAX_DEPTH = 4;
|
|
37
|
+
|
|
38
|
+
function decide(permissionDecision, reason) {
|
|
39
|
+
process.stdout.write(
|
|
40
|
+
JSON.stringify({
|
|
41
|
+
hookSpecificOutput: {
|
|
42
|
+
hookEventName: 'PreToolUse',
|
|
43
|
+
permissionDecision,
|
|
44
|
+
permissionDecisionReason: reason,
|
|
45
|
+
},
|
|
46
|
+
}),
|
|
47
|
+
);
|
|
48
|
+
}
|
|
49
|
+
|
|
50
|
+
/**
|
|
51
|
+
* Splits a command line into segments of words. Honours single and double
|
|
52
|
+
* quotes and backslash escapes; splits on unquoted && || ; | & and newlines.
|
|
53
|
+
*/
|
|
54
|
+
function segments(command) {
|
|
55
|
+
const out = [];
|
|
56
|
+
let words = [];
|
|
57
|
+
let word = '';
|
|
58
|
+
let inWord = false;
|
|
59
|
+
let quote = null;
|
|
60
|
+
const endWord = () => {
|
|
61
|
+
if (inWord) words.push(word);
|
|
62
|
+
word = '';
|
|
63
|
+
inWord = false;
|
|
64
|
+
};
|
|
65
|
+
const endSegment = () => {
|
|
66
|
+
endWord();
|
|
67
|
+
if (words.length) out.push(words);
|
|
68
|
+
words = [];
|
|
69
|
+
};
|
|
70
|
+
for (let i = 0; i < command.length; i++) {
|
|
71
|
+
const c = command[i];
|
|
72
|
+
if (quote) {
|
|
73
|
+
if (c === quote) quote = null;
|
|
74
|
+
else if (c === '\\' && quote === '"' && i + 1 < command.length) word += command[++i];
|
|
75
|
+
else word += c;
|
|
76
|
+
continue;
|
|
77
|
+
}
|
|
78
|
+
if (c === '"' || c === "'") {
|
|
79
|
+
quote = c;
|
|
80
|
+
inWord = true;
|
|
81
|
+
} else if (c === '\\' && i + 1 < command.length) {
|
|
82
|
+
word += command[++i];
|
|
83
|
+
inWord = true;
|
|
84
|
+
} else if (c === ';' || c === '|' || c === '&' || c === '\n') {
|
|
85
|
+
endSegment();
|
|
86
|
+
} else if (c === ' ' || c === '\t' || c === '(' || c === ')' || c === '{' || c === '}') {
|
|
87
|
+
endWord();
|
|
88
|
+
} else {
|
|
89
|
+
word += c;
|
|
90
|
+
inWord = true;
|
|
91
|
+
}
|
|
92
|
+
}
|
|
93
|
+
endSegment();
|
|
94
|
+
return out;
|
|
95
|
+
}
|
|
96
|
+
|
|
97
|
+
/** `$(…)` and backtick bodies, inspected as nested commands. */
|
|
98
|
+
function substitutions(command) {
|
|
99
|
+
const found = [];
|
|
100
|
+
for (const m of command.matchAll(/\$\(([^()]*)\)/g)) found.push(m[1]);
|
|
101
|
+
for (const m of command.matchAll(/`([^`]*)`/g)) found.push(m[1]);
|
|
102
|
+
return found;
|
|
103
|
+
}
|
|
104
|
+
|
|
105
|
+
function basename(word) {
|
|
106
|
+
const parts = word.split('/');
|
|
107
|
+
return parts[parts.length - 1];
|
|
108
|
+
}
|
|
109
|
+
|
|
110
|
+
/** `@scope/name@1.2.3` → `@scope/name`, `name@latest` → `name`. */
|
|
111
|
+
function packageName(spec) {
|
|
112
|
+
const at = spec.lastIndexOf('@');
|
|
113
|
+
return at > 0 ? spec.slice(0, at) : spec;
|
|
114
|
+
}
|
|
115
|
+
|
|
116
|
+
function runCodeIn(words) {
|
|
117
|
+
return words.includes('run-code');
|
|
118
|
+
}
|
|
119
|
+
|
|
120
|
+
const ASK_RUN_CODE =
|
|
121
|
+
'ask:playwright-cli run-code executes JavaScript in the Playwright node process (outside the browser sandbox); ' +
|
|
122
|
+
'confirm the code is yours and contains no text taken from page content. Set QUALIOW_HOOKS=off to disable.';
|
|
123
|
+
|
|
124
|
+
const GIT_EXEC_OPTION = /^(--upload-pack|--receive-pack|--exec|--ext-diff|--output|--config-env)(=|$)/;
|
|
125
|
+
|
|
126
|
+
/** A git option that runs a program or writes a file outside the read-only calls the skills make. */
|
|
127
|
+
function gitExecOption(words) {
|
|
128
|
+
let k = 0;
|
|
129
|
+
// Global options before the subcommand: `-c` is a config override there (and a
|
|
130
|
+
// harmless diff flag after it, as in `git log -c`).
|
|
131
|
+
while (k < words.length && words[k].startsWith('-')) {
|
|
132
|
+
if (words[k] === '-c') return '-c';
|
|
133
|
+
k += /^(-C|--git-dir|--work-tree|--namespace)$/.test(words[k]) ? 2 : 1;
|
|
134
|
+
}
|
|
135
|
+
const hit = words.slice(k).find((w) => GIT_EXEC_OPTION.test(w)) || words.slice(0, k).find((w) => GIT_EXEC_OPTION.test(w));
|
|
136
|
+
return hit ? hit.split('=')[0] : null;
|
|
137
|
+
}
|
|
138
|
+
|
|
139
|
+
/** Returns 'deny:<reason>', 'ask:<reason>' or null for one segment's words. */
|
|
140
|
+
function checkWords(words, depth) {
|
|
141
|
+
let i = 0;
|
|
142
|
+
while (i < words.length && /^[A-Za-z_][A-Za-z0-9_]*=/.test(words[i])) i++;
|
|
143
|
+
while (i < words.length && WRAPPERS.has(basename(words[i]))) {
|
|
144
|
+
i++;
|
|
145
|
+
while (i < words.length && (/^[A-Za-z_][A-Za-z0-9_]*=/.test(words[i]) || words[i].startsWith('-'))) i++;
|
|
146
|
+
}
|
|
147
|
+
if (i >= words.length) return null;
|
|
148
|
+
const program = basename(words[i]);
|
|
149
|
+
const rest = words.slice(i + 1);
|
|
150
|
+
|
|
151
|
+
if (SHELLS.has(program)) {
|
|
152
|
+
const c = rest.findIndex((w) => /^-[a-z]*c$/.test(w));
|
|
153
|
+
if (c >= 0 && rest[c + 1] != null) return checkCommand(rest[c + 1], depth + 1);
|
|
154
|
+
return null;
|
|
155
|
+
}
|
|
156
|
+
if (program === 'eval') return checkCommand(rest.join(' '), depth + 1);
|
|
157
|
+
|
|
158
|
+
if (program === 'playwright-cli') return runCodeIn(rest) ? ASK_RUN_CODE : null;
|
|
159
|
+
|
|
160
|
+
if (program === 'git') {
|
|
161
|
+
const option = gitExecOption(rest);
|
|
162
|
+
return option
|
|
163
|
+
? `ask:git ${option} can run a program or write an arbitrary file; confirm it is intended. Set QUALIOW_HOOKS=off to disable.`
|
|
164
|
+
: null;
|
|
165
|
+
}
|
|
166
|
+
|
|
167
|
+
if (program !== 'npx') return null;
|
|
168
|
+
|
|
169
|
+
let yes = false;
|
|
170
|
+
const packages = [];
|
|
171
|
+
let j = 0;
|
|
172
|
+
for (; j < rest.length; j++) {
|
|
173
|
+
const w = rest[j];
|
|
174
|
+
if (w === '--') {
|
|
175
|
+
j++;
|
|
176
|
+
break;
|
|
177
|
+
}
|
|
178
|
+
if (!w.startsWith('-')) break;
|
|
179
|
+
if (w === '-c' || w === '--call' || w.startsWith('--call=')) {
|
|
180
|
+
return 'deny:npx -c/--call runs an arbitrary shell string; call the tool directly instead. Set QUALIOW_HOOKS=off to disable.';
|
|
181
|
+
}
|
|
182
|
+
if (w === '-y' || w === '--yes' || w.startsWith('--yes=')) yes = true;
|
|
183
|
+
else if (w === '-p' || w === '--package') {
|
|
184
|
+
if (rest[j + 1] != null) packages.push(rest[++j]);
|
|
185
|
+
} else if (w.startsWith('--package=')) packages.push(w.slice('--package='.length));
|
|
186
|
+
}
|
|
187
|
+
const command = rest.slice(j);
|
|
188
|
+
const candidates = packages.length ? packages : command.length ? [command[0]] : [];
|
|
189
|
+
if (packages.length || yes) {
|
|
190
|
+
const unknown = candidates.map(packageName).filter((p) => !NPX_ALLOWED_PACKAGES.has(p));
|
|
191
|
+
if (unknown.length) {
|
|
192
|
+
return (
|
|
193
|
+
`deny:npx would install and run ${unknown.join(', ')}, which is not on the qualiow allowlist ` +
|
|
194
|
+
`(${[...NPX_ALLOWED_PACKAGES].join(', ')}). Install it deliberately instead. Set QUALIOW_HOOKS=off to disable.`
|
|
195
|
+
);
|
|
196
|
+
}
|
|
197
|
+
}
|
|
198
|
+
if (command.length) {
|
|
199
|
+
const bin = PLAYWRIGHT_CLI_NAMES.has(packageName(command[0])) ? 'playwright-cli' : basename(command[0]);
|
|
200
|
+
if (bin === 'playwright-cli' && runCodeIn(command.slice(1))) return ASK_RUN_CODE;
|
|
201
|
+
}
|
|
202
|
+
return null;
|
|
203
|
+
}
|
|
204
|
+
|
|
205
|
+
/** Worst decision across every segment and nested command. */
|
|
206
|
+
function checkCommand(command, depth = 0) {
|
|
207
|
+
if (typeof command !== 'string' || depth > MAX_DEPTH) return null;
|
|
208
|
+
let ask = null;
|
|
209
|
+
const nested = substitutions(command).map((s) => checkCommand(s, depth + 1));
|
|
210
|
+
const direct = segments(command).map((w) => checkWords(w, depth));
|
|
211
|
+
for (const r of [...nested, ...direct]) {
|
|
212
|
+
if (!r) continue;
|
|
213
|
+
if (r.startsWith('deny:')) return r;
|
|
214
|
+
ask = ask || r;
|
|
215
|
+
}
|
|
216
|
+
return ask;
|
|
217
|
+
}
|
|
218
|
+
|
|
219
|
+
try {
|
|
220
|
+
if (process.env.QUALIOW_HOOKS === 'off') process.exit(0);
|
|
221
|
+
|
|
222
|
+
let raw = '';
|
|
223
|
+
try {
|
|
224
|
+
raw = readFileSync(0, 'utf-8');
|
|
225
|
+
} catch {
|
|
226
|
+
process.exit(0);
|
|
227
|
+
}
|
|
228
|
+
if (!raw || !raw.trim()) process.exit(0);
|
|
229
|
+
|
|
230
|
+
const input = JSON.parse(raw);
|
|
231
|
+
if (input.tool_name !== 'Bash') process.exit(0);
|
|
232
|
+
|
|
233
|
+
const command = (input.tool_input || {}).command;
|
|
234
|
+
if (!command || typeof command !== 'string') process.exit(0);
|
|
235
|
+
|
|
236
|
+
const result = checkCommand(command);
|
|
237
|
+
if (result && result.startsWith('deny:')) decide('deny', result.slice('deny:'.length));
|
|
238
|
+
else if (result && result.startsWith('ask:')) decide('ask', result.slice('ask:'.length));
|
|
239
|
+
process.exit(0);
|
|
240
|
+
} catch {
|
|
241
|
+
process.exit(0);
|
|
242
|
+
}
|
|
@@ -1,8 +1,11 @@
|
|
|
1
1
|
export interface SecretPattern {
|
|
2
2
|
category: string;
|
|
3
3
|
re: RegExp;
|
|
4
|
-
|
|
4
|
+
/** Receives the match and its capture groups, as `String.prototype.replace` would. */
|
|
5
|
+
filter?: (match: string, ...groups: string[]) => boolean;
|
|
5
6
|
}
|
|
6
7
|
export const SECRET_PATTERNS: SecretPattern[];
|
|
7
8
|
/** Category names of every secret-shaped pattern found in `text`; never the matched text. */
|
|
8
9
|
export function findSecretCategories(text: string): string[];
|
|
10
|
+
/** Number of secret-shaped matches per category in `text`; never the matched text. */
|
|
11
|
+
export function countSecrets(text: string): Map<string, number>;
|