[0xLLVEL]

Summary

Closes #7244.

Adds two projection operations over caller-selected Android declarations — project_android_jni_names and project_android_attack_surface — plus a scoped packed/split APK playbook (static pack detection and split acquisition only), so an agent gets expected JNI/JNA symbol names and a ranked exposure view of manifest components in one call each instead of re-deriving them from raw manifest/DEX text every session.

Problem and expected behavior

For a selected class/method, no REA operation returns the symbol names native code would resolve (Java_pkg_Class_method, wide/short overloads, DEX shorty), so agents either re-derive the strings per call or discover mismatches only at symbol-resolution time. Likewise, the manifest component inventory comes back as declarations with nothing ranking which exported components are reachable without a permission guard, nor an intent-action index / foreground-service-type / provider-authority summary for triage.

Expected behavior: two MCP/CLI operations over caller-selected facts (no APK re-read) that return deterministic, Evidence-retained projections with explicit limitations — "string projection, not a runtime check" and "static declarations, not exploitability" — and that propagate component coverage (complete/partial/unknown) instead of claiming absence.

Change and scope

• Domain (projection owners): src/domain/android/androidJniNames.ts (JNI/JNA name projection over UTF-16 code units, the JNA default-mapping name, overload groups from distinct selected descriptors, the canonical dexDescriptorSchema, an omitted descriptor normalized once to null, per-field limits and a projected-symbol budget that also counts each retained descriptor) and src/domain/android/androidAttackSurface.ts (unguarded/guarded/guard-unknown/export-unknown findings that each carry permission_source, permission_overrides and path_permissions, intent-action index, coverage propagation, component/action/aggregate-input/projected-output budgets, and a refusal of permission values without their established source). src/domain/android/androidAnalysis.ts exports the shared androidManifestComponentSchema so both projections and the landed component-inventory contract consume one schema.
• Producing boundary (raw identities and guard facts): bridge/android/ReaJadxBridge.java emits each method's frozen raw name as dex_name; src/android/JadxProtocol.ts, src/android/JadxAnalysis.ts and androidResultSchemas.inspect_android_class expose it as dex_method_name together with dex_class_name (the raw binary class name), so callers can feed real DEX identities instead of decompiler aliases. src/android/AndroidManifestComponents.ts projects permission_overrides (none/read/write/read-write) and path_permissions (none/present) per component, defaulted to unknown when re-parsing older Evidence, and records a limitation when either is present.
• Application: AndroidJniNamesService and AndroidAttackSurfaceService create Evidence (derived confidence, analyst-inference authority, explicit limitations). Both accept an optional source_evidence — the inspect_android_package or inspect_android_class result the facts came from — authenticated by AndroidProjectionSource before projection: the record's own identity is re-derived with safeParseEvidence, every selected component, class and method must appear in it, component_coverage: "complete" requires that inventory to report complete coverage and to be passed in full, package_name must match, and the JNI class_name must be the raw dex_class_name. The result copies its artifact subject, links its evidence_id and records source_evidence_id.
• Contracts and registration: two new operations wired through applicationToolContracts, toolEffects, toolFamilies, toolOutputSchemaGroups, cliCommandNames, applicationCommands, and server/registerApplicationTools/*.
• Docs and skill: docs/android-analysis.md and the source agent skill (.agents/skills/reverse-engineer-anything/references/android-applications.md, from which skills/ is generated) teach both projections and their source_evidence handoffs; docs/android-packed-analysis.md stays within static pack detection, split acquisition and the out-of-scope boundary (CONTRIBUTING.md excludes packing-defeat recovery, documentation included), linking the general Frida guide instead of a dump recipe.
• Reconciliation with landed work: the branch's duplicate manifest parser was removed during the upstream/main merges in favor of the landed inventory parser; src/contracts/toolFamilies.ts drops a dead re-export flagged by knip; a style: commit restores upstream import layout in 26 unrelated files that merge-time lint-staged had reformatted, so this diff stays scoped.
• Intentionally not included: runtime symbol resolution, exploitability claims, changes to JADX execution, and linking the projections into inspect_android_package's summary view.

Follow the contribution scope.

Contract and boundary impact

• Semantic owner and earliest changed stage: src/domain/android owns the projection semantics; earliest changed stage is validated caller-selected declarations (strict zod schemas) in the application services. Superseded path removed: the branch-local duplicate manifest parser, replaced by the landed inventory parser behind the shared exported schema.
• Contribution scope fit and linked maintainer decision: ordinary feature inside the existing Android analysis boundary — no new provider, target execution, injection, mutation, or protection recovery; no maintainer scope decision required.
• CLI and MCP/tool-catalog contract: two additive operations (project_android_jni_names, project_android_attack_surface) registered in the tool catalog; metadata:check green. Each input schema carries only the facts its projection consumes plus the optional source_evidence link: project_android_attack_surface takes package_name, components, component_coverage and source_evidence, with no unused target_sdk/permissions, and it is bounded to 1000 components, 128 actions or categories per component, about 1 MiB of selected text and about 4 MiB of projected findings. Additive fields on the landed Android result contract: inspect_android_class carries dex_class_name and per-method dex_method_name, and every projected manifest component carries permission_overrides and path_permissions.
• Provider, bridge, target-format, or platform compatibility: none — pure projections, engine-agnostic.
• Evidence, artifact, provenance, or reconstruction contract: derived confidence, analyst-inference authority, rawResult: null, limitations stating the results do not read an APK or establish exploitability; component coverage preserved as complete/partial/unknown; the attack-surface parameters retain the selected components so fields and components the result omits stay readable, and unresolved guard facts stay unknown rather than absent, including guard presence that predates the projection, which is reported as unknown rather than as a declared guard. With source_evidence, the record's own identity is re-derived before any fact is read from it, every selected component, class and method is authenticated against it, and the new Evidence copies its subject (path, SHA-256, format) and links its id, so identical declarations from different APKs stay distinguishable; without it the Evidence states that no source was linked and keeps evidence_links empty. A selection may be narrowed from the source, never extended, and a linked complete claim requires the source inventory to be complete and to be passed in full.
• Target selection, authority, upstream behavior, and cleanup details: caller-selected declarations only; no files opened, no processes started, no cleanup obligations.
• Generated metadata (docs/public/product-catalog.json), package, or installation impact: catalog is CI-generated and validated; metadata:check passes locally; no package/dependency changes in this diff.

Expected effects (check known effects; use documentation-only on its own).

• Read existing files or captured data.
• Launch or attach to a process.
• Inject instrumentation or run scripts/code.
• Change process memory, registers, files, or configuration.
• Contact network services.
• Recover protected code or work around protections.
• Documentation or usability only; no target interaction.
• Not sure yet; investigation or maintainer guidance needed.

Evidence and regression coverage

• Tests added or updated: androidJniNames.test.ts (short/long/JNA names, distinct-descriptor overload groups, UTF-16 supplementary characters, omitted versus null descriptor, method and byte budgets), androidAttackSurface.test.ts (unguarded/guarded/guard-unknown/export-unknown split, provider overrides including a provider with a base permission and an override, path-protected providers, contradictory permission provenance, action index, coverage propagation, owned note constants, and the component/text and projected-output budget rejections), AndroidJniNamesService.test.ts and AndroidAttackSurfaceService.test.ts (Evidence shape, retained selection, the source-Evidence subject and link, and refusals of facts the linked record does not report), tests/boundary/android/JadxAnalysis.test.ts (raw dex_class_name/dex_method_name beside aliases), tests/boundary/android/androidManifestComponents.test.ts (per-provider override and path-permission projection), and the real-bridge conformance regression asserts dex_name. New public coverage: tests/boundary/mcp/androidStructuralProjectionsMcp.test.ts (advertised tools, both successes, rejected input, get_evidence_bundle readback) and tests/boundary/cli/androidStructuralProjectionsCli.test.ts (compiled-CLI successes and rejection).
• Base reproduction or other evidence: new capability, not a fix; no base reproduction applies.
• User-visible CLI/MCP output: captured below from the compiled CLI.
• Remaining proof gaps: full test suite (check:pr) and a real JADX/JDK end-to-end run were not executed locally; representative tool output is not attached.

For evidence-bearing changes:

• Observed, derived, and inferred claims remain distinguishable.
• Artifact identity, source provenance, and failed attempts remain preserved.
• Unsupported, incomplete, unavailable, or uncertain outcomes remain visible.

Representative output, compiled CLI (abridged to normalized_result):

$ rea project-android-jni-names '{"class_name":"io.example.Native","methods":[{"method_name":"nativeInit","dex_descriptor":"(I)V"},{"method_name":"nativeInit"},{"method_name":"café","dex_descriptor":"()V"}]}'
  class_name: io.example.Native
  methods[3]{method_name,dex_descriptor,jni_short_name,jni_long_name,jna_expected_name,overload_group_size}:
    nativeInit,(I)V,Java_io_example_Native_nativeInit,Java_io_example_Native_nativeInit__I,nativeInit,2
    nativeInit,null,Java_io_example_Native_nativeInit,null,nativeInit,2
    café,()V,Java_io_example_Native_caf_000e9,Java_io_example_Native_caf_000e9__,café,1

$ rea project-android-attack-surface '{"package_name":"io.example","component_coverage":"partial","components":[one unguarded activity, one exported provider with projected readPermission/writePermission overrides, one exported provider with projected child path-permission guards]}'
  exported_without_permission[1]: io.example.Main
  exported_guarded: []
  exported_guard_unknown[2]:
    - io.example.Bridge | permission_overrides read-write | "Exported; provider readPermission/writePermission overrides are projected only as present, so neither their value nor the absence of an operation guard is established."
    - io.example.Paths  | path_permissions present      | "Exported; child <path-permission> guards cover only the paths they declare and their values are not projected, so the provider's access guard is not established."
  coverage: partial
  unknowns[5]:
    - Permission declarations do not establish protection levels or runtime enforcement.
    - Reachability through intents, tasks, and deep links was not traced; listed triggers are static declarations.
    - 1 exported provider(s) declare android:readPermission or android:writePermission; those override values are not projected, so they stay guard-unknown even where a base permission is declared.
    - 1 exported provider(s) declare child <path-permission> guards; their paths and permission values are not projected, so those providers stay guard-unknown.
    - Component inventory coverage is partial; absence of a finding is unknown.

Validation performed

• npm run typecheck — green on the final tree, before and after each commit.
• npx vitest run src/domain/android src/application/android tests/boundary/android — 10 files, 230 tests passed (4 files skipped on this host), including the raw-identity, override, path-guard, provenance-link, source-identity and budget regressions.
• npm run test:focused -- tests/boundary/mcp/androidStructuralProjectionsMcp.test.ts tests/boundary/cli/androidStructuralProjectionsCli.test.ts — 4 passed against freshly rebuilt artifacts, so both operations are exercised through the public MCP and compiled-CLI boundaries.
• npx vitest run tests/boundary/mcp/compactInputSchemaMcp.test.ts — 14 tests passed, so the two input schemas (including the optional source_evidence) still fit the advertised compact profile and its per-tool budget.
• npx vitest run src/contracts/toolEffects.test.ts tests/conformance/contracts/mcpToolFamilies.test.ts tests/acceptance/tools/mobileGraphExamples.test.ts tests/conformance/readmeConsistency.test.ts — 34 tests passed.
• The focused runs above rebuilt the runtime and packaged skill from the current tree (compile, artifacts:skills) before executing, and npm run docs:check is green.
• npx vitest run src/domain/analysisView — 53 files, 1229 tests passed after the component and class result schemas gained fields.
• node scripts/ci/static-check-report.mjs lint — exit 0 (warnings only), run with --ignore-pattern ".worktrees/**"; npm run verify:module-boundaries (3144 files), npm run knip, npm run verify:architecture-guards and npm run docs:check all green, and oxfmt --check is clean on the changed files.
• Hook limitation: the pre-push npm run check:fast cannot run in this checkout because a stale scratch worktree (.worktrees/pr-3026-merge, left from a closed PR) carries a .oxlintrc.json that repo-wide oxlint rejects (reportUnusedDisableDirectives outside the root config). The equivalent checks were run manually as listed above; the push used HUSKY=0 for that environmental reason only.
• Pre-existing local failures on Windows: tests/boundary/cli/jsonInputStatus.test.ts fails three format variants because the asserted temp path is JSON-escaped in stdout (the two rows added for these commands pass); tests/boundary/mcp/androidLateCompletionMcp.test.ts and three tests/boundary/process/androidAnalysisMcp.test.ts rows fail because the JADX provider fixture is unavailable on this host. The late-completion failure reproduces at ef948eed6 without the raw-identity changes, so it is environmental, not a regression.
• Real-provider verification gap: the bridge change is asserted by the existing real-android-metadata conformance regression (real JADX + APK), which cannot run in this environment; no real-provider run was performed here.
• Earlier merge-reconciliation runs (npm ci for the zod upgrade, npm run check, focused batches) are unchanged and still green on this tree.

Compatibility, safety, and release

• Breaking changes or migration steps: none in the final diff. The branch's earlier feat(android)! contract change was absorbed into the landed component-inventory contract during reconciliation; inspect_android_package output matches main, and the two new operations are additive.
• Real Hopper/Ghidra, browser, or OS coverage: not applicable — no provider, OS, or process behavior changed; the projections are pure functions.
• Package or release metadata impact: none.
• Security, privacy, process, or containment review: no execution, no network, no file writes; inputs are strictly validated by zod schemas, permission provenance is admitted only in consistent pairs, a linked source has its own identity re-derived before any fact is read and is checked against the selection rather than asserted, coverage claims are bounded by what the linked inventory establishes, and both projections bound the selected text and the projected representation before allocating it, so one call cannot expand a modest request into an unbounded response or Evidence payload.

Author readiness

Human owner: @0xLLVEL

• I reviewed the final diff and relevant existing code, understand the change and its edge cases, and take responsibility for the complete PR, including AI-assisted work.
• I checked the evidence above, confirmed it supports the current change, and recorded meaningful validation gaps.
• I declare this PR ready for initial maintainer review.

Before merge

• This PR addresses a concrete problem or an agreed enhancement.
• Material scope and authority questions are resolved, and any required maintainer scope decision covers the final diff.
• The PR has one focused outcome and the title follows type(scope): outcome.
• Related issue is linked, or the reason for not linking one is stated above.
• Tests cover changed observable behavior and meaningful failure paths.
• Owning docs, contracts, and generated metadata are updated where needed.
• User-visible CLI/MCP changes include representative output.
• I checked the final diff for secrets, unrelated cleanup, and unsupported claims.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-10-27T11:46:55.294873Z We couldn’t complete this review. Comment @codex review to try again.	1b8ec17	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[greptile-apps]

Confidence Score: 4/5

[Medium impact] The PR adds bounded static Android projections to the CLI and MCP surface.

Not ready to merge because inline source Evidence still disappears from retained bundles.

Findings

1.  Inline sources disappear from bundles ▶
2.  Reported authorities are rejected ▶

Fix with agent prompt

### Issue 1
src/server/registerApplicationTools/projectAndroidJniNames.ts:38-42
When `source_evidence` is an inline record not already saved in this session, both new handlers save only the derived Evidence. The services return the source ID but not the source record, and `recordResult` does not follow links. `get_evidence_bundle` and `export_evidence_bundle` therefore omit the inspection behind that link, and later source lookups fail.

Return the authenticated source and save it before the derived record, using the existing source-retention path. `AGENTS.md` requires preserving caller-selected inputs and local Evidence. The same issue exists in `projectAndroidAttackSurface.ts`.

### Issue 2
src/domain/android/androidAttackSurface.ts:51-53
The new `.max(65_535)` rejects manifest facts that `inspect_android_package` already reports, even when they fit both aggregate budgets. The parser's capacity regression retains a 900,000-character provider authority, but this schema refuses that provider on its own. Splitting the selection cannot shorten one field, and editing it fails the linked-source comparison.

The stated JVM constant-pool limit does not govern manifest authority strings. Align these field bounds with the producing parser and the aggregate budgets. `AGENTS.md` requires limits to follow real format or resource constraints; this requirement must be satisfied before merging.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Summary

Adds static Android projections for expected JNI/JNA names and manifest exposure facts.

• Android analysis projects expected JNI and JNA names from selected DEX methods.
• Android package declarations become a ranked view of exported components and their triggers.
• Projected Android results can retain a checked link to their source Evidence.
• The Android guide explains static pack detection and full split acquisition.
• Small JavaScript and native-analysis cleanup leaves behavior unchanged.

This review checked source code only. No runtime tests were run. PR supersession could not be checked because GitHub access was unavailable.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Caller-selected declarations] --> B[Validate selection and budgets]
  S[Optional source Evidence] --> C[Check source ID and selected facts]
  B --> C
  C --> D[Project expected names or manifest findings]
  D --> E[Create derived Evidence]
  E --> F[CLI output]
  E --> G[MCP retention and delivery]

Loading

_{Reviews (11) · Last reviewed commit: "Merge remote-tracking branch 'upstream/m..." · Reviewed by Greptile}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9ac0b79ce1

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[0xLLVEL]

Both reviews are addressed in ef948eed6. Each inline thread has a reply and is resolved; the substantive changes:

Structural fixes

• Findings now carry permission_source. exported_without_permission requires an established permission_source: "none" on a non-provider component; unresolved permissions and exported providers whose guard may come from readPermission/writePermission (never projected by the inventory) go to a new exported_guard_unknown bucket with a matching unknowns entry.
• parameters retain the selected components and component_coverage, so dropped fields and non-exported components stay readable and calls that differ only in declarations get distinct Evidence.
• target_sdk and permissions are gone from the attack-surface input: neither affected the projection and exported_default already carries the SDK rule.
• The JNI projection reuses the canonical dexDescriptorSchema, normalizes an omitted descriptor once to null, and mangles UTF-16 code units so supplementary characters match javac -h. The class_name description now advertises pkg.Outer$Inner, and the method ceiling is the named JNI_NAME_METHOD_LIMIT with stated recovery.
• Descriptor guidance in the tool description, docs/android-analysis.md and the no-descriptor unknown now point at the dex_descriptor that inspect_android_class reports.
• docs/android-packed-analysis.md is narrowed to static pack detection and split acquisition: the dump-and-reanalyze sequence is removed (which also removes the broken start-adb-app Frida step), the CONTRIBUTING.md boundary is stated, and the general Frida guide is linked instead.

Coverage added

• tests/boundary/mcp/androidStructuralProjectionsMcp.test.ts: advertised tools, both operations, an omitted descriptor, rejected input, and get_evidence_bundle readback.
• tests/boundary/cli/androidStructuralProjectionsCli.test.ts: both compiled-CLI successes plus a rejection; both commands also added to the malformed-JSON rows in jsonInputStatus.test.ts.

Local validation is recorded in the PR body. That Author readiness still needs a human owner handle; it is intentionally left for the PR owner.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ef948eed6e

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[0xLLVEL]

Second review round addressed in a764f381c; all seven threads have replies and are resolved.

• inspect_android_class now reports raw DEX identities (dex_class_name, per-method dex_method_name) taken from the bridge's frozen raw names, so the JNI projection no longer depends on decompiler aliases or dotted nested names.
• The component inventory projects per-provider permission_overrides, so a provider with neither a base permission nor readPermission/writePermission is reported unguarded again while overridden ones stay guard-unknown.
• overload_group_size counts distinct selected descriptors and the result says the group is selection-scoped; the JNI input is bounded by per-field (65535) and aggregate (1 MiB) text budgets.
• The packed playbook now separates manifest component coverage from code coverage and tells callers to pass the inspected components and status.
• Tests compare owned note constants instead of frozen sentences.

Validation and the remaining proof gaps (real JADX bridge lane, pre-existing Windows fixture failures) are recorded in the PR body.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a764f381c3

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3c691a5f76

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 150697e608

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3388e4c97b

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.

[greptile-apps]

Inline sources disappear from bundles

When source_evidence is an inline record not already saved in this session, both new handlers save only the derived Evidence. The services return the source ID but not the source record, and recordResult does not follow links. get_evidence_bundle and export_evidence_bundle therefore omit the inspection behind that link, and later source lookups fail.

Return the authenticated source and save it before the derived record, using the existing source-retention path. AGENTS.md requires preserving caller-selected inputs and local Evidence. The same issue exists in projectAndroidAttackSurface.ts.

Context Used: AGENTS.md (source)

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/server/registerApplicationTools/projectAndroidJniNames.ts
Line: 38-42

Comment:
**Inline sources disappear from bundles**

When `source_evidence` is an inline record not already saved in this session, both new handlers save only the derived Evidence. The services return the source ID but not the source record, and `recordResult` does not follow links. `get_evidence_bundle` and `export_evidence_bundle` therefore omit the inspection behind that link, and later source lookups fail.

Return the authenticated source and save it before the derived record, using the existing source-retention path. `AGENTS.md` requires preserving caller-selected inputs and local Evidence. The same issue exists in `projectAndroidAttackSurface.ts`.

**Context Used:** AGENTS.md ([source](https://github.com/morluto/rea/blob/main/AGENTS.md))

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

[greptile-apps]

Reported authorities are rejected

The new .max(65_535) rejects manifest facts that inspect_android_package already reports, even when they fit both aggregate budgets. The parser's capacity regression retains a 900,000-character provider authority, but this schema refuses that provider on its own. Splitting the selection cannot shorten one field, and editing it fails the linked-source comparison.

The stated JVM constant-pool limit does not govern manifest authority strings. Align these field bounds with the producing parser and the aggregate budgets. AGENTS.md requires limits to follow real format or resource constraints; this requirement must be satisfied before merging.

Context Used: AGENTS.md (source)

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/domain/android/androidAttackSurface.ts
Line: 51-53

Comment:
**Reported authorities are rejected**

The new `.max(65_535)` rejects manifest facts that `inspect_android_package` already reports, even when they fit both aggregate budgets. The parser's capacity regression retains a 900,000-character provider authority, but this schema refuses that provider on its own. Splitting the selection cannot shorten one field, and editing it fails the linked-source comparison.

The stated JVM constant-pool limit does not govern manifest authority strings. Align these field bounds with the producing parser and the aggregate budgets. `AGENTS.md` requires limits to follow real format or resource constraints; this requirement must be satisfied before merging.

**Context Used:** AGENTS.md ([source](https://github.com/morluto/rea/blob/main/AGENTS.md))

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.