Introducing a new way to review your PRs.

Read the announcement

morluto / rea

Reverse engineer anything with agents, from app behavior down to native binaries.

View on GitHub

27 documents · Updated Nov 20, 2026

docs / ghidra-integration

Ghidra integration

The Ghidra integration gives REA one controlled path for native binary investigation. It admits a target and a Ghidra installation, prepares an isolated analysis, and returns bounded evidence for inventory, load-image, function, and annotation workflows. It keeps imported observations separate from facts that REA can verify against the original artifact.

System role

Ghidra is an analysis provider for executable targets. The TypeScript side selects the target and analysis contract. A Java bridge then runs inside one Ghidra Program and exposes bounded observations to REA.

The provider supports three sources of analysis state:

SourcePurposeState meaning
Fresh importAnalyze a private copy of the selected executable.Results are observations from the current Ghidra analysis.
Existing projectInspect a selected Ghidra project through a private copy.Stored project facts are historical evidence. They do not prove that REA ran or completed that analysis.
Owned checkpointReopen a saved private Program for supported stock imports.Restored facts are retained state. They are not new discoveries.

The provider advertises inventory and function operations as analyst capabilities. It also reports the limits of each result. A missing function, string, reference, or file offset is therefore not treated as proof of absence.

Provider boundaries

The provider keeps the original executable and a caller-selected source project unchanged. Fresh work uses a private target copy. Existing-project work uses a fingerprinted private project copy. Checkpoint work uses a private packed Program that REA owns and can restore. Windows P0 adds a native authority boundary for process ownership and private runtime paths. Installation admission makes that host decision before the provider reports availability. References: src/…/GhidraProviderCapabilities.ts• L22-57, src/…/GhidraProjectSnapshot.ts• L238-313, src/…/GhidraLauncher.ts• L333-369, src/…/GhidraAdmission.ts• L31-83

Session lifecycle

The session lifecycle turns an admitted artifact, project, or checkpoint into one authenticated analysis view. Each session binds the target digest, provider release, profile digest, language choice, analysis state, and available operations before it accepts normal investigation requests.

The main flow is:

flowchart LR
    A[Target or selected project] --> B[Installation and target admission]
    B --> C[Committed analysis profile]
    C --> D[Private copy or saved Program]
    D --> E[Headless Ghidra analysis]
    E --> F[Authenticated Java bridge]
    F --> G[Serialized request]
    G --> H[Validated REA evidence]
    H --> I[Checkpoint or cleanup]

The client starts Ghidra once for a session and serializes requests for its Program. Startup and shutdown have separate deadlines. Decompilation and native value-flow work have their own resource limits. Active request cancellation closes the owned session so unsaved analysis edits cannot leak into later work.

Session sources

Fresh import waits for default auto-analysis to complete before exposing normal results. Existing-project mode does not rerun analysis or repair the selected project. Owned-checkpoint mode restores a previously saved Program when its target, profile, manifest, and packed-file identity still match.

Checkpointing is opt-in for supported stock ELF, PE, and Mach-O imports on Linux and macOS. When enabled, normal close saves the private analysis state and later sessions can restore it. A failed save keeps the live session and its cleanup ownership for retry. Cancellation or process failure can lose changes after the last successful save. An interrupted first analysis cannot resume without a completed initial checkpoint. References: src/…/GhidraProviderCapabilities.ts• L31-36, src/…/GhidraProviderCapabilities.ts• L60-66, src/…/GhidraClient.ts• L1206-1289, src/…/GhidraClient.ts• L1402-1455

Installation admission

Installation admission makes the Ghidra runtime a known input before analysis starts. It checks the selected release, its declared Java range, a full 64-bit JDK, the headless launcher, and the native decompiler selected by the JVM host identity.

The verified build is Ghidra 12.2 with Java 25. Other 12.x releases from 12.2 onward can pass admission when their declared Java bounds and native tools match. A passing installation still carries an unverified-build limitation when it is not the verified build. Builds below 12.2 and other major releases are refused because the packaged bridge is not verified for them.

Installation checks

The check result distinguishes a missing setup from a contradictory setup. It retains the observed paths, release, Java identity, host platform, and native-tool result for setup and doctor flows. Relative installation and Java paths are rejected before Java runs.

The Java probe runs with selected JVM option variables cleared. It reports os.name and os.arch because those values control Ghidra's native-tool selection. The native decompiler is checked by its executable header instead of by its directory name. This catches a tool or JVM built for a different host while leaving an unreadable header as an unknown diagnostic. Java application properties are read with Java Properties rules, and malformed escapes make the installation unavailable. References: src/…/GhidraInstallation.ts• L100-175, src/…/GhidraInstallation.ts• L244-281, src/…/GhidraInstallation.ts• L888-1022, src/…/GhidraJavaProperties.ts• L12-48

Release policy

Release policy separates the verified build from the admitted release line. Ghidra 12.2 is the verified build. Later compatible 12.x builds are admitted with an unverified-build limitation and retain their observed version in provider identity.

Admitted builds also carry known provider defects when their decompiler behavior can affect evidence. The current policy records the pre-12.2.1 Clang x86-64 switch defect and the open AArch64 masked-byte switch defect. These are limitations on affected investigations, not evidence that the target itself is malformed. References: src/…/GhidraReleasePolicy.ts• L1-11, src/…/GhidraReleasePolicy.ts• L42-121, src/…/GhidraReleasePolicy.ts• L123-175

Admission decision

The admission decision combines installation status, the selected source mode, and Windows native readiness into the provider's one availability result. Existing-project mode is rejected on Windows before the project is opened. A Windows installation is accepted only when its native authority is available, so doctor and provider readiness use the same host decision. References: src/…/GhidraAdmission.ts• L39-90, src/…/GhidraProvider.ts• L197-275

Doctor diagnostics

The doctor flow projects installation checks into actionable provider status. On Windows it also checks native isolation authority and the ability of the REA process to write and remove a file under the Ghidra session root. The doctor reports an available provider only when admission passes and this temporary-directory probe succeeds.

The temporary-directory probe proves only REA-side access. It does not prove that the child JVM can write there. A launcher message that names a denied java_home.save file is classified separately so restricted-host guidance stays tied to an observed failure. Native authority failures retain their stage so a missing packaged runtime is not confused with a failed native self-test. References: src/…/GhidraDoctor.ts• L32-103, src/…/GhidraDoctor.ts• L106-167

Java bridge

The Java bridge is the Ghidra-side boundary for analysis. It authenticates each request, checks the imported Program identity, applies the session's resource limits, and maps Ghidra observations into bounded JSON results.

Bridge protocol

The bridge uses one request and one response per line. A request contains an operation, parameters, request identity, and session token. A response returns the matching identity and either a result or a bounded failure. The startup handshake reports provider, run, target, profile, language, analysis state, capabilities, and preparation facts.

The bridge refuses normal work when the imported identity is unavailable or does not match the admitted target. This keeps a successful process start separate from a valid analysis session. The TypeScript side validates input before sending it and validates output before creating REA evidence. It rejects malformed envelopes, unexpected fields, invalid addresses, and replies that do not account for the requested operation. Requests above the configured UTF-8 JSON ceiling are refused before socket write. References: bridge/…/ReaGhidraBridge.java• L204-269, bridge/…/ReaGhidraBridge.java• L404-526, src/…/GhidraSessionValues.ts• L159-207, src/…/GhidraProviderClient.ts• L996-1075

Local transport

POSIX sessions use a private Unix socket. Windows sessions use an authenticated IPv4 loopback endpoint because the Windows path does not use Java Unix sockets. The endpoint record contains only the loopback address and port. The bearer token remains in the authenticated exchange.

The socket observer reads the endpoint under the session's resource owner and rejects a record that changes during observation. The connector polls until the bridge is ready. The client treats identity or protocol failures as final rather than retrying them as transport loss. References: src/…/GhidraProviderCapabilities.ts• L39-43, src/…/GhidraProviderClient.ts• L601-708, bridge/…/ReaGhidraBridge.java• L280-385

Session authority

Fresh POSIX sessions can apply temporary names and function-entry annotations. Existing-project sessions and Windows sessions are read-only. Annotation replies include per-request outcomes and explicit conflict reasons. An invalid reply leaves the mutation result unknown rather than claiming success or failure.

Annotations affect only the private analysis database. Ephemeral changes disappear on close. REA-owned checkpoint mode can retain them after a successful save, but changes after the last saved checkpoint can still be lost. References: src/…/GhidraSessionValues.ts• L16-28, src/…/GhidraProviderCapabilities.ts• L102-109, src/…/GhidraClient.ts• L1402-1455, bridge/…/ReaGhidraBridge.java• L462-555

Analysis preparation

Analysis preparation turns target facts and caller settings into one deterministic profile before Ghidra imports the target. The profile records choices that can change coordinates or meaning. Its digest binds those choices to the session and to reusable evidence.

Analysis profiles

The profile records the loader, language, compiler specification, analyzer preset, target context, evidence rules, optional extensions, seeds, and checkpoint or project state. A later handshake must report the same committed choices. A mismatch fails closed instead of mixing results from different imports.

Header-based targets can use Ghidra's loader selection. Raw binaries need a caller-supplied processor, base address, and entry point. DOS MZ and COM targets use real-mode context. GameCube DOL targets use section and BSS facts. Verified processor lanes can pin a language when stable import behavior is required. Caller language overrides are rejected when they conflict with a pinned lane. References: src/…/GhidraAnalysisProfile.ts• L71-99, src/…/GhidraAnalysisProfile.ts• L126-233, src/…/GhidraAnalysisProfile.ts• L234-380

Preparation lanes

Preparation helpers cover formats whose load image or processor rules need more context than a standard header import provides.

LaneAnalysis purpose
Raw binaryMap file bytes at the admitted base address and use the admitted entry context.
DOS MZ or COMPreserve real-mode load and register context.
GameCube DOLMap text and data sections at load addresses and represent BSS as uninitialized memory.
PSP AllegrexUse the verified loader with the required matching extension.
RISC-VPin the admitted little-endian language and supported calling-convention policy.
MIPS and PowerPCPreserve target-specific language, storage, and function-boundary rules.

DOS COM preparation mirrors the selected Ghidra loader decision. A recognized .com image uses ComLoader at 0000:0100. An image that fails the name, MZ, or recognition gates uses an explicit BinaryLoader fallback at 1000:0100, and the reason remains part of the profile. This makes the import choice reproducible instead of relying on a later loader guess. References: src/…/GhidraComImport.ts• L13-79, src/…/GhidraComImport.ts• L126-210, src/…/GhidraAnalysisProfile.ts• L261-367

The provider compares the resulting load image with independent target facts where a lane supports that check. Each comparison records the expected value, observed value, match state, and relevant file or address location. DOS MZ and COM preparation also accounts for relocation and entry-context facts. A mismatch remains visible as evidence or causes admission to fail. A successful Ghidra launch does not hide it. References: src/…/GhidraLoadImageAttest.ts• L49-81, src/…/GhidraLoadImageAttest.ts• L153-242, src/…/GhidraDosSnapshot.ts• L27-137

Analysis controls

Disabled analyzers and caller-provided seeds are part of the profile. Optional Java extensions are copied into the private runtime. Their reported identity is checked during the session handshake.

A seed file supplies function, code, or label assertions before auto-analysis. REA snapshots and hashes the canonical form, then checks the bridge report against that commitment. Seeded facts remain caller assertions. They are not reported as discoveries made by Ghidra. A changed seed file, missing report, different digest, excess outcome, or extension mismatch causes the session to fail closed. References: src/…/GhidraLauncher.ts• L370-410, src/…/GhidraAnalysisSeeds.test.ts• L96-143, src/…/GhidraProviderClient.ts• L418-433

Optional extensions are explicit local analysis packages. REA records their path, digest, entry class, and integration API before launch. The package is copied into the private runtime, and the bridge must return one matching result. The current registered extension lane supports bounded NativeAOT metadata recovery for supported x86-64 ELF and native PE targets. A refusal can be host-based or target-based. References: src/…/GhidraExtensions.ts• L31-79, src/…/GhidraExtensions.ts• L104-185, src/…/GhidraExtensions.ts• L210-289

Existing project mode

Existing-project mode supports investigation of a caller-selected Ghidra project without opening or changing the original project. REA validates the selected marker and data directory, fingerprints the selected project content, copies only that project into private runtime state, and opens the copy as read-only.

This mode does not apply fresh loader preparation, analyzer controls, seeds, repair passes, or metadata edits. Stored project reports are historical evidence. They do not prove analysis completeness and do not become fresh REA discoveries. A change between profile resolution and copying rejects the session instead of silently analyzing a different project. References: src/…/GhidraProjectSnapshot.ts• L50-120, src/…/GhidraProjectSnapshot.ts• L123-223, src/…/GhidraProviderCapabilities.ts• L45-66, src/…/GhidraAdmission.ts• L57-73

Investigation workflows

The bridge exposes investigation as inventory, load-image, function, search, and annotation workflows. The provider validates each workflow at its boundary, adds evidence locations and limits, and keeps live observations separate from reusable snapshots.

Inventory is the first pass over a Program. It covers the imported document, names, procedures, segments, defined strings, address lookups, bytes, file-offset mappings, and binary overview facts.

Search supports procedures, strings, and bytes. Literal and regular-expression modes have different resource behavior. For case-insensitive literal procedure and string searches, the bridge returns conservative candidates and REA performs the final Unicode-aware match. Byte searches report searched coverage, skipped memory, and continuation state. An empty result therefore does not hide incomplete coverage.

Only Ghidra-defined string data belongs to the string inventory. Empty strings, references, or procedures describe what this analysis resolved. They do not prove that the artifact has no such data. References: src/…/GhidraInventoryValues.ts• L38-169, src/…/GhidraInventoryValues.ts• L322-420, src/…/GhidraProviderClient.ts• L1091-1109, src/…/GhidraProviderCapabilities.ts• L68-76

Load-image inspection

Load-image inspection checks whether Ghidra's memory and source mappings match the admitted artifact and import context. It reports source identity, mappings, permissions, entry points, entry context, relocations, and format facts.

Independent verification covers DOS MZ, DOS COM, raw-binary, and GameCube DOL lanes. PE sessions can add bounded read-only NativeAOT metadata recovery, but PE load-image verification is not claimed. Address-to-file-offset results come from observed source mappings. Uninitialized, missing, or ambiguous mappings return an explicit outcome instead of a guessed offset. References: src/…/GhidraLoadImageAttest.ts• L95-135, src/…/GhidraLoadImageAttest.ts• L153-242, src/…/GhidraLoadImageValues.ts• L124-207, src/…/GhidraProviderClient.ts• L917-967

Function investigation

Function workflows support focused study of one procedure or address. They expose identity, assembly, pseudocode, instructions, callers, callees, references, cross-references, native data types, resolved call targets, jump tables, and bounded value-flow observations.

analyze_function combines these views into a function dossier. It keeps local functions, external functions, thunks, unresolved computed flows, and unavailable facets distinct. Pseudocode, assembly, recovered types, and p-code are Ghidra observations. They are not original source or runtime proof. Relationship coverage and resource limits remain explicit when a result is partial or unknown. Function dossiers that depend on decompiler or value-flow limits are reobserved live. References: src/…/GhidraFunctionValues.ts• L133-187, src/…/GhidraFunctionValues.ts• L198-344, src/…/GhidraProviderClient.ts• L1003-1089

Session annotations

Annotation workflows add analyst names and function-entry comments to the active private database. Batch naming reports each requested key and describes contested destinations without moving an existing name silently.

Annotation-dependent reads use the live session because edits invalidate immutable analysis snapshots. A checkpoint can retain the edits only after a successful save. Existing-project and Windows sessions do not expose these mutation operations. References: src/…/GhidraSessionValues.ts• L24-28, src/…/GhidraProviderCapabilities.ts• L301-370, src/…/GhidraClient.ts• L1402-1455

Support boundaries

The verified Ghidra lane uses Ghidra 12.2 and Java 25. Supported later 12.x builds carry an unverified-build limitation. The provider supports native executable investigation only. Archives and unrelated artifact kinds are outside this integration.

Host or target classSupported workflow
Linux and macOSFresh private imports for admitted executable targets. Existing-project mode supports native ELF, PE, and Mach-O programs. Owned checkpoints support stock ELF, PE, and Mach-O imports.
Windows x64 P0Experimental read-only work for admitted native x86, x86-64, ARMNT, and ARM64 PE applications or DLLs, DOS MZ images, and selected raw 6502 images.
Raw binaryCaller-supplied architecture, address base, and entry context for admitted lanes.
DOS MZ and COMReal-mode import with required loader and entry context. Explicit COM remains outside Windows P0.
GameCube DOLPOSIX PowerPC import with section and BSS mapping.
PSP AllegrexBounded Linux x64 profile with a matching installed extension.
RISC-VLittle-endian fixed-address ELF32 or ELF64 lanes.
NativeAOTOptional bounded metadata recovery for admitted x86-64 ELF or native PE cases.

Windows P0 needs matching native isolation authority, local NTFS or ReFS paths and runtimes, and Windows 10 or later. It rejects managed, non-executable, mutable, or otherwise unsafe targets and exposes no database mutation. Universal Mach-O targets must be reduced to a slice that the session can enforce. Existing-project mode is refused on Windows before a project is opened.

For the broader provider comparison, see Reverse-engineering toolchains. For other analysis products and debugger adapters, see Debugger and disassembler integrations.

Capability limits

Default auto-analysis must complete before normal operations run. Regex search uses Java Pattern behavior. Decompilation and native value-flow work use operation limits. Large replies and bridge requests are bounded. Empty reference results describe unresolved coverage rather than proof that no reference exists.

Snapshot reuse is allowed only when the target, provider, profile, producer identity, and evidence contract still match. Reusable inventory and simple query results can come from an imported analysis snapshot. Function dossiers that depend on current decompiler or value-flow limits are reobserved live instead of replayed as immutable facts. Annotation-dependent reads are also live. References: src/…/GhidraProviderCapabilities.ts• L31-36, src/…/GhidraProviderCapabilities.ts• L68-100, src/…/GhidraSnapshotCache.test.ts• L372-445, src/…/GhidraClient.ts• L1006-1025

Change checklist

When these files or sections are changed, remember to consider these:

  • Keep installation release bounds, declared Java bounds, JVM host observations, native-tool checks, Java Properties parsing, and doctor diagnostics aligned.
  • Keep the shared admission decision aligned with provider availability, existing-project host limits, and Windows native readiness diagnostics.
  • Keep Java probe ownership preparation, bounded captured output, launcher diagnostics, and cleanup retention aligned.
  • Keep session-root write probes, launcher denial classification, and restricted-host remediation aligned; an REA-side probe does not prove child-process write access.
  • Keep profile commitments and handshake reports aligned for loader context, language selection, analyzer controls, seeds, preparation state, extensions, calling-convention policies, checkpoint identity, and existing-project state.
  • Keep COM loader recognition, snapshot filename extension, fallback coordinates, profile commitment, and independent load-image attestation aligned.
  • Keep target admission, capability reporting, architecture lanes, and the Windows read-only boundary aligned when a format or architecture lane changes.
  • Keep seed canonicalization, seed digests, bridge outcome counts, extension identity, and caller-assertion limitations aligned.
  • Keep extension package paths, JAR identity, private runtime copies, registered adapter rules, and producer result coverage aligned.
  • Keep load-image check names, expected and observed values, source locations, independent attestation, relocation evidence, NativeAOT limits, and file-offset failure reasons aligned.
  • Keep function body validation, relationship coverage, unresolved-flow records, value-flow budgets, live cache policy, and profile digest revisions aligned.
  • Keep transport authentication, endpoint publication, request-size admission, response routing, deadlines, and cleanup aligned across Unix and Windows paths.
  • Keep imported target identity admission aligned between the descriptor, Ghidra Program, startup refusal, and client handshake.
  • Keep checkpoint identity, private project copying, publication ownership, save timing, cancellation behavior, and snapshot reuse aligned.

Important failure modes

TriggerConsequenceGuard
The release is below 12.2, the JDK is not a matching 64-bit full JDK, or the native decompiler contradicts the hostThe provider is unavailable before importInstallation inspection reads release properties, Java identity, JVM OS and architecture, executable headers, and host alignment.
Java Properties contains an invalid escapeInstallation identity cannot be trustedProperty parsing rejects trailing and malformed Unicode escapes.
Existing-project mode is selected on WindowsThe provider refuses the mode before opening the selected projectShared admission rejects an existing project on a Windows host.
Java process-ownership preparation fails before launchJava does not start and the typed preparation failure remains primaryThe Java probe prepares the owned inspection first and retains cleanup dependencies for retry.
The session root blocks writes or the child JVM cannot write its java_home.save pathStartup or doctor reports a restricted-host failureProbe REA access, classify only the observed launcher denial, and retain cleanup ownership.
The imported Program has no usable target identity or its digest differs from the admitted targetThe bridge reports an identity refusal and normal operations remain unavailableThe bridge checks the imported SHA-256 before serving analysis requests.
The target, selected project, extension, or checkpoint changes during capture or restoreAdmission is rejectedPrivate copies and target, project, extension, manifest, and packed-Program identities are checked.
The handshake reports a different target, profile, language, analyzer, seed, extension, or analysis stateThe session fails closed before exposing normal resultsThe TypeScript client compares committed profile facts with bridge observations.
Auto-analysis is incompletePartial results are not exposed as a normal successful sessionThe bridge reports analysis state and blocks normal operations until completion.
A seed file changes or its reported outcome differs from the commitmentSeeded analysis is rejectedREA hashes canonical seed content and checks report counts and digest.
ComLoader rejects a DOS COM image after profile resolutionA different loader could change addresses and entry contextThe profile records the exact ComLoader or BinaryLoader choice and its rejection reason.
A bridge request or launcher reply exceeds its configured output ceiling, or a reply is malformedNo request is sent, or the operation fails before evidence creationEncoded requests and projected command output are bounded. Strict schemas validate replies and mark truncation.
A load-image check does not match its independent target factThe mismatch stays visible or admission fails for the affected laneExpected and observed values are retained with their source location.
A mapping has no unique original-file offsetOffset lookup returns a typed unavailable or ambiguous resultThe provider uses source mappings instead of virtual-address arithmetic.
A search, decompile, or value-flow result reaches its limitThe result is partial, unavailable, or live-onlyOperation limits and relationship coverage remain attached to the result.
A reference list is emptyThe result can be mistaken for proof that no reference existsProvider limitations state that the list contains only references resolved in that session.
A configured extension is missing, changed, unregistered, or reports a mismatched resultThe profile is refused or the session fails closedExtension paths, JAR bytes, adapter identity, API version, digest, and result coverage are checked.
A Windows session lacks native authority or requests a mutationThe provider is unavailable or the operation is refusedWindows readiness checks run before startup and mutation capabilities are removed.
A request is cancelled after an annotation or analysis changeUnsaved private state is discardedActive cancellation closes the owned session. Later work restores the checkpoint or reimports the source.

Key files

FileWhy to read it
src/…/GhidraProvider.tsEntry point for availability, target support, profile resolution, and client creation.
src/…/GhidraAdmission.tsDefines the shared installation and host admission decision used by readiness and provider availability.
src/…/GhidraAnalysisProfile.tsDefines the target lanes and the choices that bind an analysis session.
src/…/GhidraComImport.tsExplains explicit DOS COM loader selection and its BinaryLoader fallback.
src/…/GhidraClient.tsOwns private runtime setup, startup, serialized requests, checkpoint saves, cancellation, and cleanup.
src/…/GhidraClientTypes.tsDefines session dependencies, request ceilings, and safe lifecycle diagnostics.
src/…/GhidraProviderClient.tsConnects the session to REA workflows, evidence, limitations, attestation, extension checks, and recovery.
src/…/GhidraLauncher.tsPrepares private runtime state, copies project and extension inputs, starts headless Ghidra, and retains launch failures.
src/…/GhidraInstallation.tsDefines installation discovery, Java admission, native-tool matching, and host observations.
src/…/GhidraJavaProperties.tsDefines the Java application-properties syntax accepted during installation inspection.
src/…/GhidraReleasePolicy.tsDefines the verified release, compatible release line, and known provider defects.
src/…/GhidraTargetSnapshot.tsDefines immutable target snapshots and the filename identity seen by Ghidra loaders.
src/…/GhidraDosSnapshot.tsDefines bounded DOS source capture and independent MZ load-image attestation.
src/…/GhidraLoadImageAttest.ts and src/…/GhidraLoadImageValues.tsExplain which import lanes receive independent load-image checks and how mismatches remain visible.
src/…/GhidraExtensions.tsDefines optional extension admission, private copying, producer identity, and coverage validation.
bridge/…/ReaGhidraBridge.javaDefines the Ghidra-side identity admission, handshake, operations, transport server, resource limits, and mutation boundary.
src/…/GhidraProviderCapabilities.tsDescribes supported operations, source modes, host limits, and uncertainty rules.
src/…/GhidraAnalysisSeeds.tsDefines caller seed admission, commitment, reporting, and failure handling.
src/…/GhidraFunctionValues.tsDefines function request and result contracts, dossier evidence, and live-analysis limits.
src/…/GhidraInventoryValues.ts and src/…/GhidraAddressNameValues.tsDefine inventory, search, load-image, and naming contracts at the provider boundary.
src/…/GhidraTransport.ts, src/…/GhidraSocketConnection.ts, and src/…/GhidraClientWire.tsExplain endpoint discovery, local authentication, request limits, and reply correlation.
src/…/GhidraDoctor.ts and src/…/GhidraTemporaryDirectory.tsExplain installation readiness and restricted-host diagnostics.
src/…/GhidraAnalysisSeeds.test.ts, src/…/GhidraComImport.test.ts, and src/…/GhidraReleasePolicy.test.tsShow the guarantees around seed commitments, DOS COM import choice, and release admission.
src/…/GhidraLoadImageValues.test.ts and src/…/GhidraLoadImageAttest.test.tsShow load-image preparation, independent checks, and retained partial evidence.
src/…/GhidraExtensions.test.tsShows extension artifact admission, private copying, and producer coverage rules.
src/…/GhidraSnapshotCache.test.tsShows when cached evidence is replayed and when live analysis is required.