Command-line interface
The command-line interface (CLI) gives people and scripts one-shot access to REA investigation workflows. It turns terminal input into a checked request, runs the selected analysis or observation, and returns a result on standard output while keeping logs and progress on standard error.
CLI entrypoints
Executable entrypoint
The published rea and rea-agents commands enter through scripts/rea.mjs. This process boundary prepares the environment, checks the compiled runtime, and selects one-shot CLI mode, MCP mode, MCP HTTP mode, MCP doctor mode, or shell-completion mode. If the required runtime is absent, it reports the setup action and does not start a partial command.
flowchart LR
A[rea or npx rea-agents] --> B[scripts/rea.mjs]
B --> C{Invocation mode}
C -->|one-shot command| D[CLI workflow]
C -->|MCP| E[MCP service]
C -->|doctor| F[Readiness report]
C -->|completion| G[Shell completion data]
D --> H[Result on stdout]
References: package.json• L41-47 scripts/rea.mjs• L1-18 scripts/rea.mjs• L21-80 scripts/rea.mjs• L99-160
Command dispatch
The one-shot CLI builds one command tree and registers workflows for setup, guided workflows, analysis, artifacts, browser observation, runtime observation, evidence, environments, device access, and external analysis tools. Command families are loaded through a central registry in a stable catalog order. Family modules can load heavier services only when their commands run. Canonical command names provide the public vocabulary. Aliases support compatibility without creating a second dispatch path.
The command boundary also protects built-in MCP operations that the underlying CLI library handles separately. It normalizes valued mcp add options, checks the registration result, and turns an unknown MCP subcommand into the selected output format instead of passing a misleading success or parser message to the caller.
When a command runs, the CLI follows this sequence:
- It snapshots the selected process environment and creates the shared analysis and logging context.
- It selects the registered command and checks its arguments and options.
- It maps the checked values to the owning application workflow.
- It projects the workflow result and sets a failure exit status when the operation fails.
The CLI rejects extra positional values instead of dropping them. Completion uses the same command tree as dispatch and adds shell-specific candidate encoding. It handles completion scripts through completions <shell>. It keeps MCP doctor options separate from ordinary command options. guided-workflows provides a provider-free way to list or render existing guided prompts for callers that need help choosing a workflow. The command registry also applies common result handling after all command families are registered.
References: src/cli.ts• L109-167 src/cli.ts• L169-245 src/cli.ts• L253-264 src/cli.ts• L375-609 src/…/cliCommandRegistrations.ts• L7-17 src/…/cliCommandRegistrations.ts• L182-212 src/…/cliCommandRegistrations.ts• L264-373 src/…/cliCommandRegistrations.ts• L432-506 src/…/strictPositionals.ts• L42-109 src/…/completions.ts• L64-113 src/…/completions.ts• L115-200 src/…/guidedWorkflowCommands.ts• L10-66 src/…/mcpBuiltinCommands.ts• L18-131 src/…/projectCliAnalysisResult.ts• L9-17
Machine-readable command discovery
The CLI publishes its command vocabulary and input contracts for tools that cannot rely on terminal help. --llms gives a compact command index, --llms-full gives each command's arguments and options, and <command> --schema gives JSON Schema for that command and its global options. These documents use a versioned manifest shape so callers can discover the interface without executing a workflow.
References: src/…/commandManifests.ts• L29-55 src/…/commandManifests.ts• L67-99 src/…/commandManifests.ts• L101-142 src/…/commandManifests.ts• L144-194
Command execution model
Argument and input handling
Commands use positional arguments for the main target and named options for scope, limits, providers, selectors, and output policy. Contract projections define how shared application inputs appear as CLI arguments, including renamed fields, JSON values, caller-relative paths, branch selectors, and delivery controls. Each command checks its input before it opens a target or starts a provider. Shared schemas keep related CLI and application requests consistent.
Numeric options accept decimal command-line tokens while preserving the domain field's bounds, defaults, nullability, and help text. Published schemas describe the token form and its safe numeric range without changing the domain contract. The same schemas continue to admit JSON numbers, so command-line callers and schema-based callers use one domain contract. Native target options also bind format, architecture, entry point, provider, and optional snapshot selection to the shared target-opening request.
JSON input may be inline, read from standard input with -, or read from one local file. Named local paths inside CLI JSON are resolved from the operator's working directory. URLs, blank values, and application-relative source selections keep their own meaning for later validation. Standard input has size and memory limits, and cancellation stops the read.
Commands that accept explicit sensitive values remove those literals from parser and diagnostic text. The error code and field path remain available so a script can act on the failure without exposing the secret.
References: src/…/options.ts• L6-41 src/…/options.ts• L43-140 src/…/options.ts• L143-203 src/…/decimalTokenPattern.ts• L36-82 src/…/contractCliDescription.ts• L145-207 src/…/contractCliDescription.ts• L644-733 src/cliJsonInput.ts• L30-95 src/cliJsonInput.ts• L127-245 src/cliJsonInput.ts• L267-313 src/…/explicitSensitiveArguments.ts• L91-129
Analysis lifecycle
Most analysis commands share one one-shot lifecycle. The command validates input, passes cancellation to the workflow, retains useful partial observations, closes its provider or resource scope, and then returns the final result. A cleanup failure is kept with the main failure so a caller can see whether the result is safe to trust.
This lifecycle also covers commands that call external tools or inspect live targets. Some JavaScript analysis resources remain parked for reuse during the process, but the launcher closes them when dispatch ends. A signal becomes a cancellation request, and the CLI waits for analysis and cleanup before it reports the final status.
References: src/…/finishAnalysisCommand.ts• L17-36 src/…/finishAnalysisCommand.ts• L39-162 src/…/javascriptApplicationAnalysis.ts• L52-85 src/…/javascriptApplicationAnalysis.ts• L89-150 src/…/commandCancellation.ts• L8-31 src/…/contractCommand.ts• L662-735
Evidence delivery
Analysis commands that support Evidence can return the complete record or a smaller summary. A summary requires --evidence_output, which writes the complete source Evidence to a new local bundle. This keeps later inspection tied to the original record instead of reopening the target.
The same delivery boundary supports selected views for artifact, native, Android, browser, network-capture, and JavaScript workflows. inspect-evidence-bundle reads a saved bundle for discovery, selected records, or link membership. inspect-analysis-view projects a supported view from inline or retained Evidence and can read the exact source identity from a local bundle. It validates the requested view before publication and keeps source Evidence and derived output together when a bundle is written.
References: src/…/analysisEvidenceDelivery.ts• L25-35 src/…/analysisEvidenceDelivery.ts• L50-110 src/…/analysisEvidenceDelivery.ts• L112-163 src/…/analysisEvidenceDelivery.ts• L166-233 src/…/analysisViewCommands.ts• L22-123 src/…/contractCommand.ts• L235-277 src/…/contractCommand.ts• L314-357
Result output
Successful command data goes to standard output. The terminal-oriented default can be changed with --json or --format; supported document formats include JSON, JSON Lines, YAML, TOON, Markdown, CSV, and TSV. --filter-output selects a result subtree, --full-output adds a success or failure envelope with command metadata, and token controls support bounded human output or token counts.
The result boundary wraps registered command results while preserving framework-owned documents and generator streams. It can retain SDK output when the selected format or token control needs a final rewrite, and it frames command failures as structured output rather than allowing an earlier formatter to hide the failure details. A reusable command registry can begin a new invocation after the prior destination writes finish.
CSV and TSV are for an explicitly selected row collection. The caller must provide one row-array filter and a unique --columns list. These rules stop nested Evidence or an ambiguous result from being silently flattened. JSON Lines also needs one supported record collection.
The output boundary validates valued flags before dispatch and accepts both spaced and equals forms. It preserves structured failures, emits an empty document when a structured filter selects no value, and refuses token windows that would split structured output. Large JSON results can stream without first forming one large formatted string. TOON output stages large documents in a private temporary file before it publishes them, so a late encoding or cleanup failure does not leave a misleading prefix. Markdown and delimited output escape unsafe text, reject values that cannot be preserved in the requested encoding, and refuse spreadsheet-formula cells in CSV or TSV. A formatter failure before writing can become a typed resource error with a narrower-output remedy.
References: src/cliOutput.ts• L18-81 src/cliOutput.ts• L113-173 src/cliOutput.ts• L265-381 src/…/outputEnvelope.ts• L39-167 src/…/cliResultOutput.ts• L48-217 src/…/cliResultOutput.ts• L219-308 src/…/cliResultOutput.ts• L310-454 src/…/cliResultOutput.ts• L502-731 src/…/cliOutputFormatter.ts• L36-119 src/…/cliOutputFormatter.ts• L121-275 src/…/cliOutputFormatter.ts• L307-355 src/…/jsonLines.ts• L8-45 src/…/jsonLines.ts• L211-293 src/…/toonOutput.ts• L41-87 src/…/toonOutput.ts• L97-227
Environment setup
The CLI takes a snapshot of the selected process environment when it is created. Provider selection, setup checks, paths, logging, and other process settings use that stable view. Windows environment keys are normalized so case variants do not create different settings.
REA_LOG_LEVEL controls operational log detail. Provider preferences and provider-specific settings may also come from the environment, with an explicit command option taking precedence where supported. setup configures agent integrations and optional providers. doctor checks readiness and reports invalid configuration instead of hiding it behind normal command startup.
References: src/cli.ts• L90-127 src/…/snapshotEnvironment.ts• L1-33 src/…/options.ts• L23-88 src/…/setupCommands.ts• L34-149 src/…/setupCommands.ts• L152-204 src/…/setupCommands.ts• L219-338 src/logger.ts• L4-12 scripts/rea.mjs• L129-149
Operational logging
Command status logging
Each registered operation is measured from start to finish and emits a structured status with its command name, duration, and ok or error result. A returned failure projection counts as a failed operation even when the command returns useful error data. Unexpected errors remain process errors after the CLI records the failed operation.
Logs are JSON records written to standard error. This keeps result output valid for shell pipelines and protocol consumers. Set REA_LOG_LEVEL=silent to suppress operational logs, or use a more detailed level when diagnosing a run. --timings adds a measured command record to standard error without changing the result document.
References: src/cliLogging.ts• L82-147 src/logger.ts• L4-12 src/cli.ts• L156-167 scripts/rea.mjs• L81-97
Progress and interactive output
Interactive setup prompts and workflow progress use standard error so standard output remains a result channel. Setup can guide a person through approval and agent selection. Structured or non-interactive setup returns data for a script instead. Long browser, JavaScript, runtime, and desktop capture operations can emit newline-delimited progress records without changing their final result.
Signals map to cancellation and conventional exit codes: SIGINT maps to 130 and SIGTERM maps to 143. A reader that closes a result pipe produces the SIGPIPE status 141 rather than turning the closed pipe into a false analysis failure.
References: src/…/setupCommands.ts• L219-253 src/…/interactiveSetup.ts• L28-39 src/…/interactiveSetup.ts• L71-90 src/…/javascriptApplicationAnalysis.ts• L99-105 src/…/desktopCaptureCommands.ts• L526-594 src/…/webRuntimeCommands.ts• L17-22 src/…/commandCancellation.ts• L8-53 scripts/rea.mjs• L159-209 scripts/rea.mjs• L251-275
Command workflow families
Analysis and artifact workflows
Core commands inspect applications and binaries through overviews, decompilation, function and instruction inspection, searches, references, traces, memory views, and provider selection. Artifact and format-specific commands extend the same boundary to managed and native metadata, Android and firmware inputs, UEFI structures, Rust dependency records, packet and signal captures, SQLite, WASM, EVM, Flutter, models, GPU data, and other static sources.
These commands share the same public pattern: validate a caller-selected target, run one bounded operation, return observations or Evidence, close owned resources, and format the result. Artifact commands can also extract supported contents, decode app metadata, inspect keyed archives, trace library resolution, or inspect an ESP flash image with caller-selected address information. Firmware commands inspect candidate regions or extract selected bytes into verified files with provenance. UEFI commands map firmware volumes, files, sections, and NVRAM, or extract a selected image region with its source and provenance. Rust inspection reads embedded dependency records without executing the target. Export commands such as decompile-all, source recovery, and script export also create verified files and return a manifest, so the destination and overwrite rules are part of their contract.
The same boundary supports offline inventory of local OpenAPI declarations and saved Playwright trace archives. These commands preserve a complete source record for later selected views. Desktop application inspection identifies runtime candidates from bounded local observations, while ADB commands observe or operate on a caller-selected Android device through bounded device requests such as package, file, process, display, and log access.
References: src/…/coreBinaryCommands.ts• L51-66 src/…/coreBinaryCommands.ts• L96-188 src/…/coreNativeCommands.ts• L39-101 src/…/artifactCommands.ts• L32-81 src/…/artifactCommands.ts• L84-160 src/…/artifactCommands.ts• L163-291 src/…/artifactCommands.ts• L294-337 src/…/firmwareCommands.ts• L17-110 src/…/uefiCommands.ts• L19-66 src/…/uefiCommands.ts• L69-156 src/…/rustCommands.ts• L13-56 src/…/binaryDiagnosticsCommands.ts• L43-83 src/…/decompileDirectoryCommand.ts• L17-118 src/…/adbCommands.ts• L45-243 src/…/adbCommands.ts• L245-298 src/…/androidCommands.ts• L25-63 src/…/apktoolCommands.ts• L17-91 src/…/wasmCommands.ts• L20-92 src/…/sourceToBundleCommand.ts• L15-77 src/…/apiContractCommand.ts• L41-107 src/…/playwrightTraceCommand.ts• L38-106 src/…/desktopApplicationCommands.ts• L17-65
Browser workflows
Browser commands connect to a selected loopback browser endpoint, identify a target, and observe it within an explicit origin scope. They support passive page inspection, bundle analysis, session observation, WebMCP discovery, screenshots, and capture comparison. They are observation workflows rather than a general interactive browser shell.
Browser commands return typed failures through a common CLI error shape and can return complete or selected Evidence. See CLI browser workflows for the command-level flow and browser runtime relationship.
References: src/…/browserCommands.ts• L38-47 src/…/browserCommands.ts• L49-94 src/…/browserCommands.ts• L97-151 src/…/browserCommands.ts• L154-239 src/…/browserAdvancedCommands.ts• L41-103 src/…/browserAdvancedCommands.ts• L106-270 src/…/captureCommands.ts• L19-124
Runtime and recovery workflows
Runtime commands list and passively observe Node or Electron Inspector targets. JavaScript and Electron commands reconstruct local applications, capture owned scenarios, recover readable source files, and reconcile static results with runtime observations. Native, Frida, Cutter, GDB, Rizin, JEB, and related commands expose bounded operations against an existing process or an ephemeral session.
Reverse-engineering commands also provide one-shot objdump and Rizin inspection plus ephemeral GDB and Rizin debugger operations. These commands retain useful tool output as Evidence where supported, reject debugger actions that could break the verified output route, and close their short-lived sessions before returning. Cleanup failures remain visible instead of being treated as successful observation.
These workflows use the normal input, cancellation, cleanup, result, and failure boundaries. The CLI does not imply that a live session remains open after a one-shot command unless the command explicitly serves or controls an existing session. Desktop capture adds a checkpoint-based path for live Linux applications. It reads checkpoint requests from standard input, records bounded application state into a capture folder, and can then describe changes, inspect the captured state, or export an agent environment from selected checkpoints. See CLI evidence and runtime commands for Evidence and runtime command details.
References: src/…/javascriptRuntimeCommands.ts• L37-132 src/…/electronCommands.ts• L36-52 src/…/electronCommands.ts• L98-168 src/…/javascriptApplicationAnalysis.ts• L88-150 src/…/javascriptRecoveryCommands.ts• L13-72 src/…/desktopCaptureCommands.ts• L132-200 src/…/desktopCaptureCommands.ts• L203-280 src/…/desktopCaptureCommands.ts• L283-342 src/…/desktopCaptureCommands.ts• L526-594 src/…/fridaCommands.ts• L26-36 src/…/cutterCommands.ts• L14-87 src/…/reverseEngineeringCommands.ts• L46-180 src/…/reverseEngineeringCommands.ts• L183-480
Evidence and environment workflows
Evidence commands inspect saved bundles, import or compare records, and project selected views from complete Evidence. This lets a caller preserve a full source record while requesting a smaller report later. Input JSON commands use the same local-file and standard-input conventions as analysis commands.
Environment commands build, export, compare, and serve a local agent environment from capture Evidence. Build can inspect a HAR or read capture Evidence, then derives a manifest and state or reports what cannot be derived. Serve authenticates its manifest and capture before binding a streamable HTTP listener, then runs until cancellation or listener failure. Replay-only mode serves retained records without deriving state, which gives callers a deterministic fallback when live state is unavailable.
References: src/…/analysisViewCommands.ts• L22-123 src/…/environmentCommands.ts• L127-249 src/…/environmentCommands.ts• L273-305 src/…/environmentCommands.ts• L378-467 src/…/environmentCommands.ts• L596-709
Change checklist
When these files or sections are changed, remember to consider these:
- Keep the published
binnames, executable mode checks, compiled runtime paths, MCP routing, doctor routing, and completion routing aligned. - Keep canonical command names, aliases, registration, command schemas, shell completion, guided workflows, machine-readable manifests, and any generated command catalog aligned.
- Keep special MCP built-in parsing, output adaptation, registration verification, settings backup, and unknown-subcommand guidance aligned with the underlying CLI library.
- Preserve the separation of result output on stdout from logs, progress, prompts, and timing records on stderr.
- When changing output controls, check structured formats, JSON Lines collection admission, CSV/TSV row selection and cell safety, filters, token windows, empty results, failure envelopes, TOON staging, and SDK-owned documents together.
- When changing JSON input or numeric CLI fields, check inline JSON, stdin, files, caller-relative paths, URL handling, schema input views, decimal range publication, size limits, memory limits, and cancellation together.
- When changing environment settings, check the creation-time snapshot, Windows key handling, invalid-setting behavior, provider selection, setup, doctor, and long-running environment serving together.
- When changing Evidence delivery, check summary requirements, bundle publication, selected views, retained source Evidence, cleanup, redaction, and partial observations together.
- When changing artifact, UEFI, firmware, Rust, or native target selection, check path resolution, integrity policy, extraction destinations, format hints, flash offsets, byte ranges, output limits, and provenance together.
- When changing ADB or desktop application commands, check caller-relative paths, device or launch target selection, bounded observation limits, output directories, checkpoint input, and resource cleanup together.
- When changing offline API contract or Playwright trace inspection, check complete Evidence retention, selected-view admission, local path resolution, and bundle publication together.
- When changing reverse-engineering commands, check tool output capture, command timeouts, cancellation, debugger TTY routing, forbidden one-shot debugger actions, session cleanup, and retained process diagnostics together.
- When changing failure or cleanup classification, keep returned error data, failure logs, progress, sensitive-value redaction, and process exit status consistent.
Important failure modes
| Trigger | Consequence | Guard |
|---|---|---|
| Compiled CLI runtime is absent | The command cannot start | Build with npm ci && npm run build:cached, or reinstall the package. |
Invalid REA_LOG_LEVEL | Normal commands stop before dispatch | Use a supported level or remove the setting. Use doctor to inspect configuration. |
| Surplus positional arguments | The request is rejected instead of silently changing meaning | Check the command's declared argument shape and use --help. |
Summary delivery omits --evidence_output | The request returns an input error and cannot publish complete source Evidence | Provide a new bundle path with --detail summary. |
| JSON Lines has no single supported collection | The result is not emitted as JSON Lines | Use complete JSON or select one collection with --filter-output. |
| CSV/TSV lacks one row-array filter or columns | The result is rejected instead of being flattened ambiguously | Use --filter-output <row-array-path> and --columns <column,...>. |
| CSV/TSV contains a spreadsheet formula or an unpaired surrogate | The result is rejected rather than producing unsafe or lossy delimited text | Use JSON, or change the selected fields and values. |
| Markdown or TOON cannot preserve selected text | The result is rejected rather than silently changing the value | Use --format json or --format yaml to retain exact data. |
| Structured format is combined with a token window | The CLI rejects the output request before analysis | Use --filter-output, command pagination, or a human-readable format. |
| A provider or worker cannot close | The result can include cleanup failure and the process exits non-zero | Review cleanup details before trusting the observation. Retry after the resource is released. |
| Logs or prompts are sent to stdout | Downstream JSON or protocol consumers can fail | Keep operational text on stderr. |
| A reader closes stdout early | The writer can receive EPIPE and report the wrong pipeline status | Treat the closed pipe as SIGPIPE status 141 and stop further writes. |
| Firmware extraction uses an incomplete byte range or an unavailable output parent | No verified extraction is published | Supply both range values when selecting a range and create the output parent first. |
| A one-shot debugger command can invalidate the inferior output route | The command is rejected or cleanup can remain incomplete | Use the persistent debugger workflow for unsafe commands and inspect retained process diagnostics. |
capture-desktop-application receives no valid checkpoint request or exceeds its capture bounds | The capture cannot produce the expected checkpoint state or retains incomplete coverage | Send valid checkpoint requests and set limits that cover the intended session. |
| An ADB target is unavailable or a device-side operation fails | The requested device observation or change returns a typed failure | Confirm the serial and device state before the operation, then inspect the returned operation details. |
UEFI extraction names several offsets or uses --source-stream without one offset | The extraction request is rejected before it produces a file | Select one target offset and use --source-stream only with that target. |
| MCP registration input is swallowed by the built-in parser or its settings write is not verified | The caller can believe an agent is registered when its options or server entry were not saved | Use the CLI's normalized mcp add form and treat a non-zero or verification failure as unsuccessful. |
Key files
| File | Why to read it |
|---|---|
| scripts/rea.mjs | Shows the published process boundary, mode dispatch, output setup, cleanup, and pipe handling. |
| package.json | Defines the published command names and build entrypoints. |
| src/cli.ts | Shows CLI construction, environment capture, command registration, completion, special MCP registration handling, validation, and result wrapping. |
| src/…/cliCommandRegistrations.ts | Shows the command-family registry and the stable registration order. |
| src/cliCommandNames.ts | Defines the stable command vocabulary and aliases. |
| src/…/strictPositionals.ts | Shows how the CLI prevents unbound positional arguments. |
| src/…/guidedWorkflowCommands.ts | Shows the provider-free guided workflow entrypoint. |
| src/…/commandManifests.ts | Shows the machine-readable command index, full manifest, and JSON Schema documents. |
| src/…/contractCliDescription.ts | Defines how application contracts become CLI arguments, options, schemas, and delivery controls. |
| src/…/contractCommand.ts | Shows contract admission, routing, target opening, cancellation, and workflow execution. |
| src/…/options.ts | Defines shared CLI schema adaptation for decimal values, published input schemas, and native target opening. |
| src/…/decimalTokenPattern.ts | Defines safe decimal range publication for numeric command-line tokens. |
| src/…/help.ts | Shows terminal help layout and global option presentation. |
| src/cliOutput.ts | Defines global output formats and early output validation. |
| src/…/cliResultOutput.ts | Defines result framing, filtering, streaming, token controls, and structured failure output. |
| src/…/outputEnvelope.ts | Defines dispatcher-owned failure documents and terminal call-to-action rendering. |
| src/…/outputFilter.ts | Defines the --filter-output path grammar. |
| src/…/toonOutput.ts | Defines safe TOON publication for large output and its cleanup boundary. |
| src/…/cliOutputFormatter.ts | Defines safe Markdown and CSV/TSV rendering. |
| src/…/jsonLines.ts | Defines JSON Lines collection selection and refusal rules. |
| src/cliJsonInput.ts | Defines inline, stdin, and file JSON input plus caller-relative path resolution. |
| src/…/analysisEvidenceDelivery.ts | Defines complete versus summary Evidence delivery and bundle publication. |
| src/…/analysisViewCommands.ts | Defines saved-bundle inspection and selected analysis-view projection. |
| src/…/finishAnalysisCommand.ts | Defines cancellation, resource cleanup, partial observations, and combined failures. |
| src/…/commandCancellation.ts | Defines signal ownership, cancellation propagation, and conventional exit statuses. |
| src/cliLogging.ts | Defines command status, duration, failure classification, and exit status. |
| src/…/snapshotEnvironment.ts | Defines the stable environment view used by a CLI instance. |
| src/…/setupCommands.ts | Shows setup, readiness, uninstall, and update entrypoints. |
| src/…/interactiveSetup.ts | Shows interactive setup prompts, approval, progress, and stderr use. |
| src/…/artifactCommands.ts | Shows artifact inspection, extraction, metadata decoding, and target selection. |
| src/…/firmwareCommands.ts | Shows firmware region inspection and verified byte extraction. |
| src/…/uefiCommands.ts | Shows UEFI inspection and offset-based extraction commands. |
| src/…/rustCommands.ts | Shows Rust dependency inspection and Evidence delivery. |
| src/…/reverseEngineeringCommands.ts | Shows one-shot external-tool and debugger workflows with Evidence and cleanup boundaries. |
| src/…/browserCommands.ts | Shows the main passive browser command family. |
| src/…/apiContractCommand.ts | Shows offline OpenAPI inventory and selected Evidence delivery. |
| src/…/playwrightTraceCommand.ts | Shows offline Playwright trace inventory and selected Evidence delivery. |
| src/…/desktopApplicationCommands.ts | Shows bounded desktop application inspection. |
| src/…/desktopCaptureCommands.ts | Shows desktop capture checkpoints, state inspection, task drafting, and environment export. |
| src/…/adbCommands.ts | Shows bounded Android device observation and control commands. |
| src/…/mcpBuiltinCommands.ts | Shows MCP built-in command discovery and output-safe unknown-subcommand handling. |
| src/cliEvidenceCommands.ts | Shows Evidence import, export, and comparison entrypoints. |
| src/…/environmentCommands.ts | Shows environment build, export, comparison, and serving workflows. |