morluto/rea

Fork 0

mirror of https://github.com/morluto/rea.git synced 2026-11-05 06:45:22 +00:00

[GH-ISSUE #4708] [Bug] Electron inspection turns an invalid resource size into a definite zero #2457

New issue

Closed

opened 2026-10-17 21:04:56 +00:00 by bsakaguchi · 4 comments

bsakaguchi commented

2026-10-17 21:04:56 +00:00

Owner

Originally created by @coygeek on GitHub (Oct 17, 2026).
Original GitHub issue: https://github.com/morluto/rea/issues/4708

Before submitting

I searched the open and closed issues for a related report.
I removed credentials, tokens, private paths, and sensitive target data.

Area

Process capture or runtime observation

Observed behavior

The public inspect-electron-page CLI returned content_size: 0 for an authorized local resource whose Page.getResourceTree record supplied contentSize: -1. It returned no resource exclusion or unavailable section. This reproduced in two of two calls on a fresh main build and two of two calls on the latest published npm package, using a controlled loopback HTTP/WebSocket CDP stand-in. This establishes handling of deliberately malformed metadata; no real Electron or Chromium release was observed emitting negative sizes.

The six inputs were [-1, 0, 128, omitted, "bad", null]; output sizes were [0, 0, 128, null, null, null]. The negative and actual-zero controls become indistinguishable. The supplied string and null values also receive no malformed-resource diagnostic.

Reproduction

Use the repository-pinned Node.js and npm toolchain, run npm ci, then npm run build:cached in a fresh checkout of the source revision in REA version. These steps were executed successfully for the source test.
From that repository root, run the fixture below. By default it calls the freshly built CLI. To test an installed published package instead, set the shell variable cli_path to that package's scripts/rea.mjs absolute path before running it. The same inline fixture was executed in both modes.
The fixture creates existing regular files in a disposable home, starts a loopback CDP endpoint, invokes the public CLI twice, and closes the endpoint. Its assertions describe the observed failing behavior. No real Electron application or external target is required.

verification_home=$(mktemp -d)
trap 'rm -rf "$verification_home"' EXIT
env -i PATH="$PATH" HOME="$verification_home" TMPDIR="$verification_home" LANG=C node --import tsx --input-type=module - "${cli_path:-scripts/rea.mjs}" <<'JS'
import assert from "node:assert/strict";
import { execFile } from "node:child_process";
import { writeFile } from "node:fs/promises";
import { join, basename } from "node:path";
import { pathToFileURL } from "node:url";
import { promisify } from "node:util";
const { startFakeCdpBrowser } = await import(pathToFileURL(join(process.cwd(), "tests/fixtures/fakeCdpBrowser.ts")).href);
const cliPath = process.argv[2] ?? "scripts/rea.mjs";
const sizes = [-1, 0, 128, undefined, "bad", null];
const page = join(process.env.HOME, "index.html");
await writeFile(page, "<main>fixture</main>");
await Promise.all(sizes.map((_, i) => writeFile(join(process.env.HOME, `resource-${i}.js`), "export const value = 123;")));
const url = pathToFileURL(page).href;
const browser = await startFakeCdpBrowser({
  electronFileUrl: url,
  commandEvents: () => [],
  commandResult: ({ method }) => method === "Page.getResourceTree" ? {
    frameTree: { frame: { id: "frame-main", url }, resources: sizes.map((size, i) => ({
      url: new URL(`resource-${i}.js`, url).href, type: "Script", mimeType: "text/javascript",
      ...(size === undefined ? {} : { contentSize: size })
    })) }
  } : undefined,
});
try {
  for (let run = 1; run <= 2; run++) {
    const { stdout } = await promisify(execFile)(process.execPath, [cliPath, "inspect-electron-page", browser.endpoint, "electron-page", "--observation-ms", "0", "--json"], { env: process.env });
    const result = JSON.parse(stdout);
    assert.equal(result.operation, "inspect_electron_page");
    const value = result.normalized_result;
    const resources = value.resources.map(r => ({ file: basename(r.file_path), content_size: r.content_size })).sort((a,b) => a.file.localeCompare(b.file));
    assert.deepEqual(resources.map(r => r.content_size), [0, 0, 128, null, null, null]);
    const excluded = value.completeness.excluded.filter(e => e.section === "resources");
    assert.deepEqual(excluded, []);
    assert(!value.completeness.unavailable_sections.includes("resources"));
    console.log(JSON.stringify({ run, resources, resource_exclusions: excluded, unavailable_sections: value.completeness.unavailable_sections }));
  }
  const commands = browser.commands.map(c => c.method);
  assert.equal(commands.filter(m => m === "Page.getResourceTree").length, 2);
  console.log(JSON.stringify({ resource_tree_reads: 2, detach_calls: commands.filter(m => m === "Target.detachFromTarget").length }));
} finally { await browser.close(); }
JS

Expected behavior

Preserve uncertainty when producer metadata cannot establish a nonnegative size. A finite negative value must not become a definite zero. Retain other valid resource facts with an unknown size and explicit incomplete coverage for the malformed field. An omitted optional size remains unknown without being confused with a malformed supplied value; actual zero and positive sizes retain their values.

The repository requires preservation of uncertainty and source evidence across boundaries, explicit unknowns when normalization loses information, and explicit completeness in Electron observations. Its nullable resource-size schema provides an unknown representation. The shared web parser already retains invalid supplied sizes as null with a resource diagnostic. The protocol declares an optional number and does not document a negative sentinel; this report does not claim a normal producer emits one.

Impact and scope

The observed consequence is a bounded evidence-integrity failure: callers cannot distinguish the malformed negative size from a resource actually measured as zero, and completeness does not reveal invalid metadata. All four fixture calls reproduced it. Frequency in real Electron or Chromium releases is unknown. No security boundary crossing or target mutation was established.

Impact level: P2
Assessment status: supported
Impact basis: The public CLI fabricates a valid-looking zero and omits a malformed-resource diagnostic for controlled producer metadata; impact is confined to this result's evidence integrity.
Workaround status: unverified
Workaround basis: No practical caller-side workaround preserving the original unknown size and its provenance was verified.

REA version

Fresh source build: main commit 5b24399387. Published-package test: rea-agents@6.5.0, confirmed as npm latest before installation. Both CLIs returned 6.5.0 from --version; the source build completed with locked dependencies and no restored build tasks.

Runtime and platform

Node.js v24.18.0; npm 11.16.0 (the repository's pinned toolchain); Darwin 27.0.0 arm64. Each probe used a fresh HOME and only PATH, HOME, TMPDIR and LANG in its environment. No real Electron or Chromium runtime was used.

Target and artifact context

A loopback HTTP/WebSocket CDP stand-in with a file-page target and six existing regular local JavaScript resources. Resource type and MIME fields are valid; only size metadata varies. The stand-in supplies malformed values deliberately and establishes no occurrence in a real browser. The published-package test used the same source-checkout fixture to drive its installed public CLI.

Interface

CLI

Earliest failure stage

Capture or observation

Relevant output or evidence

Expected source: AGENTS.md:17 and AGENTS.md:31; docs/electron-observation.md:57-59; src/domain/javascript/electronObservation.ts:135; corroborating shared web resource admission at src/browser/CdpCaptureDocuments.ts:120-126.
Failure source: src/browser/CdpElectronInspection.ts:298-301; the public CLI fixture above reaches this conversion after regular-file authorization.
Evidence provenance: observed
Local verification: reproduced
Reproduction completeness: complete

Both tested runtimes exited successfully and returned the following output for each pair of CLI calls (summarized without temporary paths):

{"run":1,"content_sizes":[0,0,128,null,null,null],"resource_exclusions":[],"unavailable_sections":[]}
{"run":2,"content_sizes":[0,0,128,null,null,null],"resource_exclusions":[],"unavailable_sections":[]}
{"resource_tree_reads":2,"detach_calls":2}

Source inspection separately confirms that numberValue admits finite negative values, regular-file authorization precedes conversion, Math.max clamps the size to zero without a completeness entry, and output validation accepts it. The official CDP Page.FrameResource declaration describes contentSize as an optional number with no negative-size sentinel. CLI transport and normalization were tested; real-producer incidence and MCP execution were not tested.

Typed outcome or recovery expected

With the same fixture, the negative size should remain unknown with a resource completeness diagnostic for invalid supplied metadata. Retain its valid file identity, type and MIME facts. Zero and positive controls must remain 0 and 128; omitted optional size must remain unknown without a malformed-value diagnostic. Supplied string and null controls must remain unknown with invalid-metadata coverage. Restoration must preserve these distinctions rather than discard all resources or accept negative sizes as valid.

Analysis or suspected cause

Source-derived reasoning: the Electron resource loop passes finite negative values to a clamp without informing the completeness ledger. The web path separately checks nonnegative size admission. No real-producer cause or negative sentinel is established.

Additional context

Live open/closed issue and PR searches found no existing report covering this Electron scalar failure. Issue #3379 and merged PR #3901 cover malformed Electron collections; this fixture supplies a valid resource collection and valid resource records. PR #3236 documents the analogous web resource-size problem, with its production change confined to the web resource parser; it is closed and unmerged as a PR, while the web correction is present in the tested source. Merged PR #3416 adds web CLI/MCP uncertainty coverage. These changes do not correct or exercise this Electron scalar conversion. No last-known-good release was established.

Originally created by @coygeek on GitHub (Oct 17, 2026). Original GitHub issue: https://github.com/morluto/rea/issues/4708 ### Before submitting - [X] I searched the open and closed issues for a related report. - [X] I removed credentials, tokens, private paths, and sensitive target data. ### Area Process capture or runtime observation ### Observed behavior The public `inspect-electron-page` CLI returned `content_size: 0` for an authorized local resource whose `Page.getResourceTree` record supplied `contentSize: -1`. It returned no resource exclusion or unavailable section. This reproduced in two of two calls on a fresh main build and two of two calls on the latest published npm package, using a controlled loopback HTTP/WebSocket CDP stand-in. This establishes handling of deliberately malformed metadata; no real Electron or Chromium release was observed emitting negative sizes. The six inputs were `[-1, 0, 128, omitted, "bad", null]`; output sizes were `[0, 0, 128, null, null, null]`. The negative and actual-zero controls become indistinguishable. The supplied string and null values also receive no malformed-resource diagnostic. ### Reproduction 1. Use the repository-pinned Node.js and npm toolchain, run `npm ci`, then `npm run build:cached` in a fresh checkout of the source revision in REA version. These steps were executed successfully for the source test. 2. From that repository root, run the fixture below. By default it calls the freshly built CLI. To test an installed published package instead, set the shell variable `cli_path` to that package's `scripts/rea.mjs` absolute path before running it. The same inline fixture was executed in both modes. 3. The fixture creates existing regular files in a disposable home, starts a loopback CDP endpoint, invokes the public CLI twice, and closes the endpoint. Its assertions describe the observed failing behavior. No real Electron application or external target is required. ```bash verification_home=$(mktemp -d) trap 'rm -rf "$verification_home"' EXIT env -i PATH="$PATH" HOME="$verification_home" TMPDIR="$verification_home" LANG=C node --import tsx --input-type=module - "${cli_path:-scripts/rea.mjs}" <<'JS' import assert from "node:assert/strict"; import { execFile } from "node:child_process"; import { writeFile } from "node:fs/promises"; import { join, basename } from "node:path"; import { pathToFileURL } from "node:url"; import { promisify } from "node:util"; const { startFakeCdpBrowser } = await import(pathToFileURL(join(process.cwd(), "tests/fixtures/fakeCdpBrowser.ts")).href); const cliPath = process.argv[2] ?? "scripts/rea.mjs"; const sizes = [-1, 0, 128, undefined, "bad", null]; const page = join(process.env.HOME, "index.html"); await writeFile(page, "<main>fixture</main>"); await Promise.all(sizes.map((_, i) => writeFile(join(process.env.HOME, `resource-${i}.js`), "export const value = 123;"))); const url = pathToFileURL(page).href; const browser = await startFakeCdpBrowser({ electronFileUrl: url, commandEvents: () => [], commandResult: ({ method }) => method === "Page.getResourceTree" ? { frameTree: { frame: { id: "frame-main", url }, resources: sizes.map((size, i) => ({ url: new URL(`resource-${i}.js`, url).href, type: "Script", mimeType: "text/javascript", ...(size === undefined ? {} : { contentSize: size }) })) } } : undefined, }); try { for (let run = 1; run <= 2; run++) { const { stdout } = await promisify(execFile)(process.execPath, [cliPath, "inspect-electron-page", browser.endpoint, "electron-page", "--observation-ms", "0", "--json"], { env: process.env }); const result = JSON.parse(stdout); assert.equal(result.operation, "inspect_electron_page"); const value = result.normalized_result; const resources = value.resources.map(r => ({ file: basename(r.file_path), content_size: r.content_size })).sort((a,b) => a.file.localeCompare(b.file)); assert.deepEqual(resources.map(r => r.content_size), [0, 0, 128, null, null, null]); const excluded = value.completeness.excluded.filter(e => e.section === "resources"); assert.deepEqual(excluded, []); assert(!value.completeness.unavailable_sections.includes("resources")); console.log(JSON.stringify({ run, resources, resource_exclusions: excluded, unavailable_sections: value.completeness.unavailable_sections })); } const commands = browser.commands.map(c => c.method); assert.equal(commands.filter(m => m === "Page.getResourceTree").length, 2); console.log(JSON.stringify({ resource_tree_reads: 2, detach_calls: commands.filter(m => m === "Target.detachFromTarget").length })); } finally { await browser.close(); } JS ``` ### Expected behavior Preserve uncertainty when producer metadata cannot establish a nonnegative size. A finite negative value must not become a definite zero. Retain other valid resource facts with an unknown size and explicit incomplete coverage for the malformed field. An omitted optional size remains unknown without being confused with a malformed supplied value; actual zero and positive sizes retain their values. The repository requires preservation of uncertainty and source evidence across boundaries, explicit unknowns when normalization loses information, and explicit completeness in Electron observations. Its nullable resource-size schema provides an unknown representation. The shared web parser already retains invalid supplied sizes as null with a resource diagnostic. The protocol declares an optional number and does not document a negative sentinel; this report does not claim a normal producer emits one. ### Impact and scope The observed consequence is a bounded evidence-integrity failure: callers cannot distinguish the malformed negative size from a resource actually measured as zero, and completeness does not reveal invalid metadata. All four fixture calls reproduced it. Frequency in real Electron or Chromium releases is unknown. No security boundary crossing or target mutation was established. - Impact level: P2 - Assessment status: supported - Impact basis: The public CLI fabricates a valid-looking zero and omits a malformed-resource diagnostic for controlled producer metadata; impact is confined to this result's evidence integrity. - Workaround status: unverified - Workaround basis: No practical caller-side workaround preserving the original unknown size and its provenance was verified. ### REA version Fresh source build: main commit 5b243993872c2bcc70cb00060ffb79b347321370. Published-package test: `rea-agents@6.5.0`, confirmed as npm latest before installation. Both CLIs returned `6.5.0` from `--version`; the source build completed with locked dependencies and no restored build tasks. ### Runtime and platform Node.js v24.18.0; npm 11.16.0 (the repository's pinned toolchain); Darwin 27.0.0 arm64. Each probe used a fresh HOME and only PATH, HOME, TMPDIR and LANG in its environment. No real Electron or Chromium runtime was used. ### Target and artifact context A loopback HTTP/WebSocket CDP stand-in with a file-page target and six existing regular local JavaScript resources. Resource type and MIME fields are valid; only size metadata varies. The stand-in supplies malformed values deliberately and establishes no occurrence in a real browser. The published-package test used the same source-checkout fixture to drive its installed public CLI. ### Interface CLI ### Earliest failure stage Capture or observation ### Relevant output or evidence - Expected source: AGENTS.md:17 and AGENTS.md:31; docs/electron-observation.md:57-59; src/domain/javascript/electronObservation.ts:135; corroborating shared web resource admission at src/browser/CdpCaptureDocuments.ts:120-126. - Failure source: src/browser/CdpElectronInspection.ts:298-301; the public CLI fixture above reaches this conversion after regular-file authorization. - Evidence provenance: observed - Local verification: reproduced - Reproduction completeness: complete Both tested runtimes exited successfully and returned the following output for each pair of CLI calls (summarized without temporary paths): ```json {"run":1,"content_sizes":[0,0,128,null,null,null],"resource_exclusions":[],"unavailable_sections":[]} {"run":2,"content_sizes":[0,0,128,null,null,null],"resource_exclusions":[],"unavailable_sections":[]} {"resource_tree_reads":2,"detach_calls":2} ``` Source inspection separately confirms that `numberValue` admits finite negative values, regular-file authorization precedes conversion, `Math.max` clamps the size to zero without a completeness entry, and output validation accepts it. The official CDP `Page.FrameResource` declaration describes `contentSize` as an optional number with no negative-size sentinel. CLI transport and normalization were tested; real-producer incidence and MCP execution were not tested. ### Typed outcome or recovery expected With the same fixture, the negative size should remain unknown with a resource completeness diagnostic for invalid supplied metadata. Retain its valid file identity, type and MIME facts. Zero and positive controls must remain `0` and `128`; omitted optional size must remain unknown without a malformed-value diagnostic. Supplied string and null controls must remain unknown with invalid-metadata coverage. Restoration must preserve these distinctions rather than discard all resources or accept negative sizes as valid. ### Analysis or suspected cause Source-derived reasoning: the Electron resource loop passes finite negative values to a clamp without informing the completeness ledger. The web path separately checks nonnegative size admission. No real-producer cause or negative sentinel is established. ### Additional context Live open/closed issue and PR searches found no existing report covering this Electron scalar failure. [Issue #3379](https://github.com/morluto/rea/issues/3379) and merged [PR #3901](https://github.com/morluto/rea/pull/3901) cover malformed Electron collections; this fixture supplies a valid resource collection and valid resource records. [PR #3236](https://github.com/morluto/rea/pull/3236) documents the analogous web resource-size problem, with its production change confined to the web resource parser; it is closed and unmerged as a PR, while the web correction is present in the tested source. Merged [PR #3416](https://github.com/morluto/rea/pull/3416) adds web CLI/MCP uncertainty coverage. These changes do not correct or exercise this Electron scalar conversion. No last-known-good release was established.

bsakaguchi closed this issue

2026-10-17 21:04:57 +00:00

bsakaguchi commented

2026-10-17 21:04:58 +00:00

Author

Owner

@morluto commented on GitHub (Oct 17, 2026):

Published 79d2b7189 on main.

Structural fix: web and Electron resource capture now share a typed resource-size parser. Omission returns unknown without a malformed-value diagnostic; supplied negative/string/null/nonfinite values return a tagged invalid result, retain null size and add invalid_protocol_value coverage. Other resource facts and zero/positive values remain intact. Removed Electron clamping and the duplicated web numeric interpretation.

Validation executed: the new public CLI regression failed against the baseline with [0,0,128,null,null,null]; after the fix it retained all six file identities and sizes [null,0,128,null,null,null], with three invalid-resource metadata exclusions and one final detach. 36 focused checks across three files passed (11 scalar controls, 24 web document controls, one compiled public CLI flow). check:fast and docs:check (5 checks) passed; initial static check caught a fixture array annotation, corrected before final checks. Commit/push hooks passed. Fake loopback CDP producer and real CLI/transport only; no real Electron/browser or MCP execution for the malformed reply.

Research: Context7/DeepWiki consulted and checked against the official Page PDL, where contentSize is an optional number describing resource content size; no negative-size sentinel is documented. This does not establish real-producer incidence.

Bounded equivalent cost task: completed six-resource production web projection using the shared parser. Baseline 748ac7f0d; all six complete replies byte-identical, 254 raw cl100k tokens each. Electron full-task costs were not measured.

Phase/build	Elapsed s	CPU s	Sampled process-family peak RSS bytes
First/baseline	0.046770	0.043785	47333376
First/candidate	0.055964	0.044510	56836096
Warm median/baseline	0.056711	0.041908	58933248
Warm median/candidate	0.057221	0.042817	58146816

Separate first processes and two alternating warm repeats; no true cold-cache or performance improvement claim. wait4 CPU/PPID-family RSS sampling may miss transients. Whole-agent/schema/retry, full CLI/MCP/Electron captures, live engines and large-output costs unmeasured. Heavy profiling/stress stayed stopped.

Normal rebase over independent Kuna deadline change9be493880; affected source/test hashes unchanged. Owned comparison worktree verified clean/inactive and removed normally. Related DOM parent and fallback-field audit (#4709) remains in progress and is not fixed by this commit.

@morluto commented on GitHub (Oct 17, 2026): Published `79d2b7189` on main. Structural fix: web and Electron resource capture now share a typed resource-size parser. Omission returns unknown without a malformed-value diagnostic; supplied negative/string/null/nonfinite values return a tagged invalid result, retain null size and add invalid_protocol_value coverage. Other resource facts and zero/positive values remain intact. Removed Electron clamping and the duplicated web numeric interpretation. Validation executed: the new public CLI regression failed against the baseline with [0,0,128,null,null,null]; after the fix it retained all six file identities and sizes [null,0,128,null,null,null], with three invalid-resource metadata exclusions and one final detach. 36 focused checks across three files passed (11 scalar controls, 24 web document controls, one compiled public CLI flow). check:fast and docs:check (5 checks) passed; initial static check caught a fixture array annotation, corrected before final checks. Commit/push hooks passed. Fake loopback CDP producer and real CLI/transport only; no real Electron/browser or MCP execution for the malformed reply. Research: Context7/DeepWiki consulted and checked against the [official Page PDL](https://github.com/ChromeDevTools/devtools-protocol/blob/master/pdl/domains/Page.pdl), where contentSize is an optional number describing resource content size; no negative-size sentinel is documented. This does not establish real-producer incidence. Bounded equivalent cost task: completed six-resource production web projection using the shared parser. Baseline 748ac7f0d; all six complete replies byte-identical, 254 raw cl100k tokens each. Electron full-task costs were not measured. | Phase/build | Elapsed s | CPU s | Sampled process-family peak RSS bytes | | --- | ---: | ---: | ---: | | First/baseline | 0.046770 | 0.043785 | 47333376 | | First/candidate | 0.055964 | 0.044510 | 56836096 | | Warm median/baseline | 0.056711 | 0.041908 | 58933248 | | Warm median/candidate | 0.057221 | 0.042817 | 58146816 | Separate first processes and two alternating warm repeats; no true cold-cache or performance improvement claim. wait4 CPU/PPID-family RSS sampling may miss transients. Whole-agent/schema/retry, full CLI/MCP/Electron captures, live engines and large-output costs unmeasured. Heavy profiling/stress stayed stopped. Normal rebase over independent Kuna deadline change9be493880; affected source/test hashes unchanged. Owned comparison worktree verified clean/inactive and removed normally. Related DOM parent and fallback-field audit (#4709) remains in progress and is not fixed by this commit.

bsakaguchi commented

2026-10-17 21:04:59 +00:00

Author

Owner

@XinwangTao commented on GitHub (Oct 17, 2026):

Agreed — turning an invalid contentSize: -1 into a definite 0 silently corrupts evidence, and a shared typed resource-size parser for both web and Electron is the right deduplication. One semantic question worth pinning down: in CDP Page.getResourceTree, contentSize is optional and some providers use -1 as a de-facto 'unknown' sentinel rather than a genuine negative. If that is the case here, mapping -1 to unknown (same as omission) instead of a tagged invalid value would match protocol expectations — otherwise, keeping the raw supplied value alongside the tagged result would at least make the invalid classification auditable in the output.

@XinwangTao commented on GitHub (Oct 17, 2026): Agreed — turning an invalid `contentSize: -1` into a definite `0` silently corrupts evidence, and a shared typed resource-size parser for both web and Electron is the right deduplication. One semantic question worth pinning down: in CDP `Page.getResourceTree`, `contentSize` is optional and some providers use `-1` as a de-facto 'unknown' sentinel rather than a genuine negative. If that is the case here, mapping `-1` to `unknown` (same as omission) instead of a tagged invalid value would match protocol expectations — otherwise, keeping the raw supplied value alongside the tagged result would at least make the invalid classification auditable in the output.

bsakaguchi commented

2026-10-17 21:04:59 +00:00

Author

Owner

@morluto commented on GitHub (Oct 17, 2026):

Fixed on main by 79d2b71896 (invalid resource sizes remain unknown).

Rechecked at 969aafcf2: the built public CLI regression passes against a controlled local CDP endpoint. Negative, omitted, string and null sizes remain null; zero and 128 are preserved; the three malformed values are accounted for in coverage, and the session detaches once. This is controlled protocol coverage, not verification against a real Electron/Chromium application or the published npm package.

@morluto commented on GitHub (Oct 17, 2026): Fixed on main by 79d2b718963b28d5b5ce42cda67db648827a9a23 (invalid resource sizes remain unknown). Rechecked at 969aafcf2: the built public CLI regression passes against a controlled local CDP endpoint. Negative, omitted, string and null sizes remain null; zero and 128 are preserved; the three malformed values are accounted for in coverage, and the session detaches once. This is controlled protocol coverage, not verification against a real Electron/Chromium application or the published npm package.

bsakaguchi commented

2026-10-31 00:34:12 +00:00

Author

Owner

@XinwangTao commented on GitHub (Oct 18, 2026):

The 79d2b71 adjustment matches CDP semantics — treating -1 as unknown rather than a tagged invalid value keeps the parser aligned with how providers actually signal 'not provided', and the regression evidence (negative/omitted/string/null → null, zero/128 preserved) covers the realistic surface. Looks good to me; thanks for folding in the feedback.

@XinwangTao commented on GitHub (Oct 18, 2026): The 79d2b71 adjustment matches CDP semantics — treating -1 as unknown rather than a tagged invalid value keeps the parser aligned with how providers actually signal 'not provided', and the regression evidence (negative/omitted/string/null → null, zero/128 preserved) covers the realistic surface. Looks good to me; thanks for folding in the feedback.

bsakaguchi referenced this issue

2026-10-31 01:05:05 +00:00

[PR #2457] [CLOSED] test(agent): grade notes export facts against produced evidence #5243

bsakaguchi referenced this issue

2026-10-31 01:06:23 +00:00

[PR #2801] [CLOSED] test(agent): grade Notes export facts and track fixture changes #5405