prooflane
Developer guide

Run it. Inspect it. Prove it.

Security scans and Prompt Regression from your terminal, with named checks and reports that explain the result.

Practical walkthroughs and 60 command help surfaces generated from the public CLI.

01 · Get started

Install the CLI and connect your model

Requires Node.js 24 or newer. Install the public CLI, then check which commands your installed version supports.

npm install -g @prooflane/cli
prooflane-inspector --help
prooflane-inspector prompt regression --help
Using the latest changes from this branch?

Until a release containing these changes is published, build them from your checked-out Prooflane repository:

npm run build:core
npm run build --workspace @prooflane/cli
node packages/cli/dist/prooflane.js prompt regression --help

For the commands below, replace prooflane-inspector with node packages/cli/dist/prooflane.js when using this source build. The recording scripts use the source build automatically.

In a Bash session, choose your provider and model, then load the provider key without displaying it. Use a model available to your provider account.

bash
export PROOFLANE_LLM_PROVIDER=openai
export PROOFLANE_LLM_MODEL="YOUR_MODEL_ID"
read -r -s -p "Provider API key: " PROOFLANE_LLM_API_KEY
printf '\n'
export PROOFLANE_LLM_API_KEY

The CLI also accepts anthropic as the provider and PROOFLANE_LLM_BASE_URL for a compatible custom endpoint.

02 · Security

Scan an authorized MCP target

Security scans use your Prooflane workspace's entitled hosted controls and execute their tasks through the local Runner. Load a Prooflane access token for the selected environment in the same terminal. This token is separate from your model-provider key and any target authentication.

export PROOFLANE_CONTROL_PLANE_URL="https://api.prooflane.ai"
read -r -s -p "Prooflane access token: " PROOFLANE_TOKEN
printf '\n'
export PROOFLANE_TOKEN
export MCP_TARGET_URL="http://127.0.0.1:9011/rpc"

prooflane-inspector security scan --url "$MCP_TARGET_URL" \
  --report security-report.html

Start your MCP server first and replace the example URL with your target. For staging, set PROOFLANE_CONTROL_PLANE_URL to your staging control-plane URL and use its access token. A scoped Gate CI token does not substitute for the hosted Security access token.

See the check being executed

Terminal progress names each check when it starts, then prints its result and completed count. Example output:

Security scan · 32 checks
Running · Prompt injection
RESISTED · Prompt injection · 1/32 complete
Running · Tool authorization
EXPOSED · Tool authorization · 2/32 complete

Names, counts, and results depend on the selected controls and your workspace entitlement. EXPOSED identifies reachable risk; BREACHED requires evidence that a boundary was crossed. INCONCLUSIVE means the check could not establish a result; NOT EXERCISED means it did not exercise the target.

The HTML report contains findings, verification details, severity, and timing. Raw prompts, model responses, and credentials are excluded from this Security report. With no model or provider key configured, deterministic MCP controls can still run; model-dependent checks are not exercised.

Security scan options · Autonomous campaign commands

03 · Prompt regression

Compare real model answers with trusted expectations

Prompt Regression calls your configured provider directly from the CLI. It needs a provider key, but no Prooflane hosted access token. Save this example as prompt-suite.json:

{
  "name": "Support policy",
  "minScore": 100,
  "system": "Answer in one sentence using the exact policy: Returns are accepted within 30 days.",
  "cases": [
    {
      "id": "return-window",
      "name": "Return-window policy",
      "prompt": "What is the return window?",
      "expected": "Returns are accepted within 30 days",
      "assert": "contains"
    }
  ]
}
prooflane-inspector prompt regression prompt-suite.json \
  --include-content \
  --out prompt-report.json \
  --report prompt-report.html

Understand exactly what failed

--include-content adds the system instruction, input prompt, expected answer, observed model answer, and assertion explanation to both local reports. Failed cases open expanded in the HTML report. Credential-shaped values are redacted and long fields are truncated; the report indicates when this happens.

Without this option, reports contain case names, verdicts, scores, timing, token counts, and fingerprints instead of prompt and answer text. This option changes only CLI artifacts; Inspector and pillar reporting are unchanged.

Assertion modes
ModeWhat is checked
exactCase-sensitive equality after trimming leading and trailing whitespace.
containsThe answer contains the trimmed expected phrase, ignoring letter case.
similarLexical token similarity: PASS at 0.80 or above, DRIFT at 0.50–0.79, otherwise FAIL. This is a text comparison, not an LLM judge.

READY requires the aggregate score to meet --min-score (or the suite threshold, default 100) with no execution errors. BLOCK exits with code 1. A local prompt result does not replace a governed release Gate.

Run the synthetic baseline and deliberate mismatch

From a repository checkout, run the committed sample suites:

prooflane-inspector prompt regression examples/prompt-regression-demo-baseline.json \
  --include-content --report baseline-report.html

prooflane-inspector prompt regression examples/prompt-regression-demo-regressed.json \
  --include-content --report regressed-report.html

The baseline expects a 30-day return window. The second suite deliberately expects 45 days while the system instruction still specifies 30 days. It demonstrates a failed assertion; it is synthetic test data. Provider failures appear as ERROR, not as proof of a behavioral regression.

All Prompt Regression options

04 · Live demos

Record two actual CLI runs

On macOS, the repository includes separate recording scripts for Security and Prompt Regression. Build the core and CLI first, load your credentials in the same terminal, and enable Terminal in System Settings → Privacy & Security → Screen & System Audio Recording.

./scripts/record-security-cli-demo.sh
./scripts/record-prompt-regression-cli-demo.sh

Each script checks live model connectivity before recording, executes the CLI, and opens the generated HTML report. Videos and reports are saved in a timestamped folder under artifacts/live-demo/. Recording captures the main display, including other visible windows.

The Security script defaults to Prooflane staging and launches the local mcp-security-scanner fixture. Set PROOFLANE_DEMO_SECURITY_SOURCE to that checkout (with its .venv installed), or set PROOFLANE_DEMO_SECURITY_TARGET to an already-running authorized MCP endpoint. PROOFLANE_CONTROL_PLANE_URL and PROOFLANE_LLM_MODEL override the script defaults.

05 · Command reference

Every command and option

Expand a command to see its arguments and flags. This reference is generated from the current source build.

prooflane-inspector
Usage: prooflane-inspector [options] [command]

Prooflane thin local Runner for MCP inspection and Hosted Intelligence

Options:
  -V, --version                 output the version number
  -h, --help                    display help for command

Commands:
  model                         inspect local AI model artifacts without
                                loading or executing them
  target                        connect and invoke HTTP/API, RAG, model, and
                                agent targets through the normalized local
                                adapter boundary
  connect [options] [config]
  list                          list MCP capabilities
  call [options] <tool>
  snapshot [options]
  show [options] <baseline>
  diff [options] <baseline>
  record [options]
  stop
  sessions
  automate [options] <session>
  benchmark [options] <suite>   compare complete AI deployment candidates with
                                paired evidence and benchmark-health checks
  prompt                        run local prompt assurance against a real model
  security                      run Security scans and manage local
                                authentication context
  security-auth                 inspect and preflight local Security
                                authentication without creating a hosted scan
  security-scan [options]       run a signed Hosted Intelligence Security scan
  redteam                       run signed hosted Security campaigns through
                                generic local MCP/LLM primitives
  gate [options] [config]       resolve and execute an approved governed Gate
                                or Suite
  suites                        inspect governed Suite artifacts
  runner                        register and operate a private Remote Runner
  debug
  auth
  project                       manage deterministic Prooflane project
                                resolution
  help [command]                display help for command
model
Usage: prooflane-inspector model [options] [command]

inspect local AI model artifacts without loading or executing them

Options:
  -h, --help                   display help for command

Commands:
  scan [options] <path>        static model supply-chain scan: formats, pickle,
                               archives, code, secrets, provenance, hashes,
                               licenses and reports
  security [options] <source>  model & supply chain security: artifact,
                               provenance, AI package and template/config packs
                               — static, no model execution
  help [command]               display help for command
model scan
Usage: prooflane-inspector model scan [options] <path>

static model supply-chain scan: formats, pickle, archives, code, secrets,
provenance, hashes, licenses and reports

Arguments:
  path                  model file or repository directory

Options:
  --no-recursive        scan only direct files in a directory
  --follow-symlinks     follow links only when they remain inside the artifact
                        root
  --max-files <n>       maximum inventory entries (default: 20000)
  --max-text-mb <n>     per-file text/config analysis limit in MiB (default: 2)
  --max-archive-mb <n>  bounded deep archive-analysis budget in MiB (default:
                        512)
  --threat-db <file>    local JSON SHA-256 threat database ({ hashes: { sha256:
                        {...} } })
  --json [file]         print JSON, or write it to an optional file
  --report <file>       write a self-contained HTML audit report
  --sarif <file>        write SARIF 2.1.0 for CI/code-scanning systems
  --fail-on <severity>  set exit code 1 for findings at/above:
                        critical|high|medium|low (default: "high")
  --quiet               suppress progress and human summary
  -h, --help            display help for command
model security
Usage: prooflane-inspector model security [options] <source>

model & supply chain security: artifact, provenance, AI package and
template/config packs — static, no model execution

Arguments:
  source                    local path, https URL, or Hugging Face repository
                            id (owner/model)

Options:
  --source-type <type>      file | directory | url | huggingface (inferred when
                            omitted)
  --revision <ref>          repository revision; a commit SHA pins immutably, a
                            branch or tag does not
  --expect-sha256 <digest>  expected artifact digest, to detect hash drift
  --packs <ids>             comma-separated packs
                            (model-artifact,model-provenance,ai-package-integrity,template-config-integrity)
  --follow-symlinks         follow links only when they remain inside the
                            artifact root
  --max-files <n>           maximum inventory entries (default: 20000)
  --max-text-mb <n>         per-file text/config analysis limit in MiB
                            (default: 2)
  --max-archive-mb <n>      bounded deep archive-analysis budget in MiB
                            (default: 512)
  --json [file]             print JSON, or write it to an optional file
  --report <file>           write a self-contained HTML audit report
  --sarif <file>            write SARIF 2.1.0 for CI/code-scanning systems
  --fail-on <severity>      set exit code 1 for findings at/above:
                            critical|high|medium|low (default: "high")
  --quiet                   suppress progress and human summary
  -h, --help                display help for command
target
Usage: prooflane-inspector target [options] [command]

connect and invoke HTTP/API, RAG, model, and agent targets through the
normalized local adapter boundary

Options:
  -h, --help                     display help for command

Commands:
  import [options] <manifest>    validate or import the canonical Prooflane
                                 connection manifest
  connect [options] <config>     test, discover, fingerprint, and save a local
                                 universal target reference
  inspect [options]              reconnect to the active universal target and
                                 inspect its normalized capability surface
  invoke [options] <capability>  run a local, non-governed preview through the
                                 active target adapter
  disconnect                     clear the active universal target reference
                                 without deleting governed history
  help [command]                 display help for command
target import
Usage: prooflane-inspector target import [options] <manifest>

validate or import the canonical Prooflane connection manifest

Arguments:
  manifest       prooflane.connections.json

Options:
  --check        validate and preview without resolving secrets, executing
                 commands, or creating connections
  --trust-stdio  explicitly trust and execute imported MCP stdio commands
  -h, --help     display help for command
target connect
Usage: prooflane-inspector target connect [options] <config>

test, discover, fingerprint, and save a local universal target reference

Arguments:
  config      local target JSON; credential values must reference
              environment-variable names

Options:
  --json      print the safe normalized target descriptor as JSON
  -h, --help  display help for command

Examples:
  prooflane-inspector target connect examples/targets/petstore-openapi.json
  PETSTORE_API_KEY=special-key prooflane-inspector target connect examples/targets/petstore-manual-auth.json
  prooflane-inspector target connect examples/targets/wikipedia-retriever.json
  GROQ_API_KEY=... prooflane-inspector target connect examples/targets/groq-model.json
  GROQ_API_KEY=... prooflane-inspector target connect examples/targets/groq-agent.json
target inspect
Usage: prooflane-inspector target inspect [options]

reconnect to the active universal target and inspect its normalized capability
surface

Options:
  --json      print the safe normalized surface as JSON
  -h, --help  display help for command

Example:
  prooflane-inspector target inspect --json
target invoke
Usage: prooflane-inspector target invoke [options] <capability>

run a local, non-governed preview through the active target adapter

Arguments:
  capability      normalized capability name or ID

Options:
  --input <file>  JSON input file
  --args <json>   inline JSON input
  -h, --help      display help for command

Examples:
  prooflane-inspector target invoke getOrderById --args '{"orderId":1}'
  prooflane-inspector target invoke retrieve --args '{"query":"OpenAI"}'
  prooflane-inspector target invoke model_inference --args '{"prompt":"Reply OK"}'
  prooflane-inspector target invoke agent_invoke --args '{"input":"Reply OK"}'
target disconnect
Usage: prooflane-inspector target disconnect [options]

clear the active universal target reference without deleting governed history

Options:
  -h, --help  display help for command
connect
Usage: prooflane-inspector connect [options] [config]

Arguments:
  config               MCP config JSON or remote URL

Options:
  --url <url>
  -e, --env <pair...>
  --cwd <dir>
  --baseline <file>
  --oauth              OAuth 2.1 + PKCE for a remote MCP server
  -h, --help           display help for command
list
Usage: prooflane-inspector list [options]

list MCP capabilities

Options:
  -h, --help  display help for command
call
Usage: prooflane-inspector call [options] <tool>

Options:
  --input <file>
  --args <json>
  --expect <file>
  -h, --help       display help for command
snapshot
Usage: prooflane-inspector snapshot [options]

Options:
  --tool <name>
  --input <file>
  --out <file>    contract path (default: "baseline.kawach.json")
  -h, --help      display help for command
show
Usage: prooflane-inspector show [options] <baseline>

Options:
  --tool <name>
  -h, --help     display help for command
diff
Usage: prooflane-inspector diff [options] <baseline>

Options:
  --tool <name>
  -h, --help     display help for command
record
Usage: prooflane-inspector record [options]

Options:
  --label <label>
  -h, --help       display help for command
stop
Usage: prooflane-inspector stop [options]

Options:
  -h, --help  display help for command
sessions
Usage: prooflane-inspector sessions [options]

Options:
  -h, --help  display help for command
automate
Usage: prooflane-inspector automate [options] <session>

Options:
  --format <format>
  --out <file>
  -h, --help         display help for command
benchmark
Usage: prooflane-inspector benchmark [options] <suite>

compare complete AI deployment candidates with paired evidence and
benchmark-health checks

Arguments:
  suite              benchmark JSON containing candidates[] and cases[]

Options:
  --trials <n>       measured repetitions
  --warmup <n>       discarded warmups
  --concurrency <n>  concurrency per candidate
  --api-key <key>    shared provider key; candidate route.keyEnv takes
                     precedence
  --out <file>       write the complete benchmark JSON report
  --json             print the complete report
  --explain          print a richer summary derived from the same completed
                     benchmark report
  -h, --help         display help for command
prompt
Usage: prooflane-inspector prompt [options] [command]

run local prompt assurance against a real model

Options:
  -h, --help                    display help for command

Commands:
  regression [options] <suite>  execute a committed prompt suite against its
                                trusted golden answers
  help [command]                display help for command
prompt regression
Usage: prooflane-inspector prompt regression [options] <suite>

execute a committed prompt suite against its trusted golden answers

Arguments:
  suite                      JSON suite containing cases[] (or legacy
                             prompts[]) with expected golden answers

Options:
  --llm-provider <provider>  openai | anthropic (or PROOFLANE_LLM_PROVIDER)
  --llm-model <model>        target model (or PROOFLANE_LLM_MODEL)
  --llm-api-key <key>        session-only provider key; PROOFLANE_LLM_API_KEY
                             is preferred
  --llm-base-url <url>       provider base URL (or PROOFLANE_LLM_BASE_URL)
  --min-score <n>            minimum score required for READY; defaults to the
                             suite value or 100
  --concurrency <n>          parallel model calls (default 4, max 16) (default:
                             4)
  --max-tokens <n>           maximum output tokens per case (default 1024)
                             (default: 1024)
  --timeout-seconds <n>      timeout per case (default 120, max 300) (default:
                             120)
  --out <file>               write a privacy-safe JSON evidence report
  --report <file>            write a standalone privacy-safe HTML report
  --include-content          include sanitized local prompt, golden-answer, and
                             model-output evidence in reports
  --json                     print the complete privacy-safe report
  -h, --help                 display help for command
security
Usage: prooflane-inspector security [options] [command]

run Security scans and manage local authentication context

Options:
  -h, --help               display help for command

Commands:
  auth                     inspect and preflight local Security authentication
                           without exposing credentials
  scan [options] [config]  run the entitled signed Security controls with
                           immutable local principal bindings
  help [command]           display help for command
security auth
Usage: prooflane-inspector security auth [options] [command]

inspect and preflight local Security authentication without exposing
credentials

Options:
  -h, --help                      display help for command

Commands:
  profiles [options]              list safe metadata for saved local Security
                                  authentication profiles
  inspect [options] <profile-id>  inspect one profile's safe metadata; secret
                                  references and values are omitted
  test [options] <profile-id>     run local authentication preflight without
                                  creating a hosted scan or reserving a run
                                  allowance
  help [command]                  display help for command
security auth profiles
Usage: prooflane-inspector security auth profiles [options]

list safe metadata for saved local Security authentication profiles

Options:
  --profiles-file <path>  local Security authentication profile registry
  --json
  -h, --help              display help for command
security auth inspect
Usage: prooflane-inspector security auth inspect [options] <profile-id>

inspect one profile's safe metadata; secret references and values are omitted

Options:
  --profiles-file <path>  local Security authentication profile registry
  --json
  -h, --help              display help for command
security auth test
Usage: prooflane-inspector security auth test [options] <profile-id>

run local authentication preflight without creating a hosted scan or reserving
a run allowance

Options:
  --profiles-file <path>  local Security authentication profile registry
  --path <path>           protected target path (defaults to the connected
                          target's tested request)
  --json
  -h, --help              display help for command
security scan
Usage: prooflane-inspector security scan [options] [config]

run the entitled signed Security controls with immutable local principal
bindings

Arguments:
  config                            MCP config JSON or remote URL

Options:
  --url <url>                       remote MCP server URL
  --bearer <token>                  MCP bearer token (kept only on the local
                                    Runner)
  --oauth                           OAuth 2.1 + PKCE for the remote MCP server
  --security-auth-preflight [path]  preflight selected target authentication
                                    before hosted scan creation
  --principal <slot=profile-id>     bind an immutable local Security principal;
                                    repeat for comparison/admin, or use
                                    anonymous=none (default: [])
  --profiles-file <path>            local Security authentication profile
                                    registry
  -e, --env <pair...>               local MCP environment override
  --cwd <dir>                       local MCP working directory
  --model <provider:model>          local target LLM; compatibility alias
  --llm-provider <provider>
  --llm-model <model>
  --llm-api-key <key>
  --llm-base-url <url>
  --prooflane-token <token>
  --prooflane-url <url>
  --report <file>                   write a standalone privacy-safe HTML graph
  --json
  -h, --help                        display help for command
security-auth
Usage: prooflane-inspector security-auth [options] [command]

inspect and preflight local Security authentication without creating a hosted
scan

Options:
  -h, --help                    display help for command

Commands:
  inspect [options] [config]
  preflight [options] [config]
  help [command]                display help for command
security-auth inspect
Usage: prooflane-inspector security-auth inspect [options] [config]

Arguments:
  config            MCP config JSON or remote URL

Options:
  --url <url>       remote MCP server URL
  --bearer <token>  local bearer token (never printed)
  --json
  -h, --help        display help for command
security-auth preflight
Usage: prooflane-inspector security-auth preflight [options] [config]

Arguments:
  config               MCP config JSON or remote URL

Options:
  --url <url>          remote MCP server URL
  --bearer <token>     local bearer token
  --oauth              OAuth 2.1 + PKCE
  --path <path>        protected target path (default: "/")
  -e, --env <pair...>  local MCP environment override
  --cwd <dir>          local MCP working directory
  --json
  -h, --help           display help for command
security-scan
Usage: prooflane-inspector security-scan [options]

run a signed Hosted Intelligence Security scan

Options:
  --target <name>            model/agent display name (default: "CLI agent")
  --control <id...>
  --prooflane-token <token>
  --prooflane-url <url>
  --prooflane-app-url <url>
  --llm-provider <provider>
  --llm-model <model>
  --llm-api-key <key>
  --llm-base-url <url>
  --json
  -h, --help                 display help for command
redteam
Usage: prooflane-inspector redteam [options] [command]

run signed hosted Security campaigns through generic local MCP/LLM primitives

Options:
  -h, --help                        display help for command

Commands:
  autonomous                        control the authorized autonomous agent
                                    boundary campaign through the local
                                    Inspector safety kernel
  campaigns [options]               list entitled hosted campaign metadata
  campaign [options] <id> [config]  run one hosted campaign end to end and
                                    optionally write its HTML graph
  scan [options] [config]           run the entitled hosted Security control
                                    set against one MCP target
  help [command]                    display help for command
redteam autonomous
Usage: prooflane-inspector redteam autonomous [options] [command]

control the authorized autonomous agent boundary campaign through the local
Inspector safety kernel

Options:
  -h, --help                     display help for command

Commands:
  status [options]               show kill-switch, connected target, discovered
                                 resources, and prior campaigns
  init [options] [file]          write an explicit campaign scope template from
                                 the currently discovered target
  start [options] <scope>        start an authorized adaptive campaign and wait
                                 for its terminal evidence state
  show [options] <campaign-id>   show one durable campaign snapshot
  watch [options] <campaign-id>  wait for one campaign to reach a terminal
                                 evidence state
  stop [options] <campaign-id>   request an early safe stop; no new campaign
                                 actions will be admitted
  help [command]                 display help for command
redteam autonomous status
Usage: prooflane-inspector redteam autonomous status [options]

show kill-switch, connected target, discovered resources, and prior campaigns

Options:
  --json
  --inspector-url <url>      loopback URL of the running local Inspector server
                             (or PROOFLANE_INSPECTOR_URL)
  --prooflane-token <token>  Prooflane access token; saved login or
                             PROOFLANE_TOKEN is preferred
  -h, --help                 display help for command
redteam autonomous init
Usage: prooflane-inspector redteam autonomous init [options] [file]

write an explicit campaign scope template from the currently discovered target

Arguments:
  file                       scope file to create (default:
                             "autonomous-boundary-scope.json")

Options:
  --force                    replace an existing scope file
  --inspector-url <url>      loopback URL of the running local Inspector server
                             (or PROOFLANE_INSPECTOR_URL)
  --prooflane-token <token>  Prooflane access token; saved login or
                             PROOFLANE_TOKEN is preferred
  -h, --help                 display help for command
redteam autonomous start
Usage: prooflane-inspector redteam autonomous start [options] <scope>

start an authorized adaptive campaign and wait for its terminal evidence state

Arguments:
  scope                      explicit JSON scope created by `redteam autonomous
                             init`

Options:
  --authorized-by <name>     operator confirming authorization for this exact
                             target and scope
  --confirm-authorized       confirm authorization for the exact target,
                             identities, resources, duration, budgets, and
                             impact boundary
  --idempotency-key <key>    stable non-sensitive retry key
  --llm-provider <provider>
  --llm-model <model>
  --llm-api-key <key>        session-only BYOK key; PROOFLANE_LLM_API_KEY is
                             preferred
  --llm-base-url <url>
  --poll-ms <milliseconds>   campaign status poll interval (default: "1500")
  --no-wait                  return after the server accepts the campaign
  --json
  --inspector-url <url>      loopback URL of the running local Inspector server
                             (or PROOFLANE_INSPECTOR_URL)
  --prooflane-token <token>  Prooflane access token; saved login or
                             PROOFLANE_TOKEN is preferred
  -h, --help                 display help for command
redteam autonomous show
Usage: prooflane-inspector redteam autonomous show [options] <campaign-id>

show one durable campaign snapshot

Options:
  --json
  --inspector-url <url>      loopback URL of the running local Inspector server
                             (or PROOFLANE_INSPECTOR_URL)
  --prooflane-token <token>  Prooflane access token; saved login or
                             PROOFLANE_TOKEN is preferred
  -h, --help                 display help for command
redteam autonomous watch
Usage: prooflane-inspector redteam autonomous watch [options] <campaign-id>

wait for one campaign to reach a terminal evidence state

Options:
  --poll-ms <milliseconds>   campaign status poll interval (default: "1500")
  --json
  --inspector-url <url>      loopback URL of the running local Inspector server
                             (or PROOFLANE_INSPECTOR_URL)
  --prooflane-token <token>  Prooflane access token; saved login or
                             PROOFLANE_TOKEN is preferred
  -h, --help                 display help for command
redteam autonomous stop
Usage: prooflane-inspector redteam autonomous stop [options] <campaign-id>

request an early safe stop; no new campaign actions will be admitted

Options:
  --json
  --inspector-url <url>      loopback URL of the running local Inspector server
                             (or PROOFLANE_INSPECTOR_URL)
  --prooflane-token <token>  Prooflane access token; saved login or
                             PROOFLANE_TOKEN is preferred
  -h, --help                 display help for command
redteam campaigns
Usage: prooflane-inspector redteam campaigns [options]

list entitled hosted campaign metadata

Options:
  --prooflane-token <token>
  --prooflane-url <url>
  --json
  -h, --help                 display help for command
redteam campaign
Usage: prooflane-inspector redteam campaign [options] <id> [config]

run one hosted campaign end to end and optionally write its HTML graph

Arguments:
  id                                hosted campaign id, for example
                                    sandbox-breakout
  config                            MCP config JSON or remote URL

Options:
  --url <url>                       remote MCP server URL
  --bearer <token>                  MCP bearer token (kept only on the local
                                    Runner)
  --oauth                           OAuth 2.1 + PKCE for the remote MCP server
  --security-auth-preflight [path]  preflight selected target authentication
                                    before hosted scan creation
  --principal <slot=profile-id>     bind an immutable local Security principal;
                                    repeat for comparison/admin, or use
                                    anonymous=none (default: [])
  --profiles-file <path>            local Security authentication profile
                                    registry
  -e, --env <pair...>               local MCP environment override
  --cwd <dir>                       local MCP working directory
  --model <provider:model>          local target LLM; compatibility alias
  --llm-provider <provider>
  --llm-model <model>
  --llm-api-key <key>
  --llm-base-url <url>
  --prooflane-token <token>
  --prooflane-url <url>
  --report <file>                   write a standalone privacy-safe HTML graph
  --json
  -h, --help                        display help for command

Example:
  prooflane-inspector redteam campaign sandbox-breakout --url https://huggingface.co/mcp --report graph.html
redteam scan
Usage: prooflane-inspector redteam scan [options] [config]

run the entitled hosted Security control set against one MCP target

Arguments:
  config                            MCP config JSON or remote URL

Options:
  --url <url>                       remote MCP server URL
  --bearer <token>                  MCP bearer token (kept only on the local
                                    Runner)
  --oauth                           OAuth 2.1 + PKCE for the remote MCP server
  --security-auth-preflight [path]  preflight selected target authentication
                                    before hosted scan creation
  --principal <slot=profile-id>     bind an immutable local Security principal;
                                    repeat for comparison/admin, or use
                                    anonymous=none (default: [])
  --profiles-file <path>            local Security authentication profile
                                    registry
  -e, --env <pair...>               local MCP environment override
  --cwd <dir>                       local MCP working directory
  --model <provider:model>          local target LLM; compatibility alias
  --llm-provider <provider>
  --llm-model <model>
  --llm-api-key <key>
  --llm-base-url <url>
  --prooflane-token <token>
  --prooflane-url <url>
  --report <file>                   write a standalone privacy-safe HTML graph
  --json
  -h, --help                        display help for command
gate
Usage: prooflane-inspector gate [options] [config]

resolve and execute an approved governed Gate or Suite

Arguments:
  config                     local MCP config JSON or remote URL

Options:
  --control <id...>
  --prooflane-token <token>
  --prooflane-url <url>
  --prooflane-app-url <url>
  --llm-provider <provider>
  --llm-model <model>
  --llm-api-key <key>
  --llm-base-url <url>
  --json
  --target <id>              connection id in prooflane.connections.json
  --trust-stdio              explicitly trust an imported MCP stdio command
  --gate <slug>              approved Gate slug or exact slug@version
  --suite <slug-or-version>  approved Suite slug or exact slug@version
  --suite-file <path>        explicit migrated legacy local Suite source;
                             resolves its matching approved Cloud Suite
  --project <id>             Prooflane project id
  --tags-file <path>         local tagged-test file (default:
                             ".prooflane/tags.json")
  --cache-dir <path>         owner-only immutable verified-artifact cache
  --offline                  execute an exact cached Gate or Suite version
                             without Control Plane access
  --project-root <dir>       repository root for project identity and local
                             tagged tests
  --url <url>                local remote MCP URL
  -e, --env <pair...>        local MCP environment override
  --cwd <dir>                local MCP working directory
  --github                   write a sanitized GitHub Step Summary and
                             annotations
  -h, --help                 display help for command
suites
Usage: prooflane-inspector suites [options] [command]

inspect governed Suite artifacts

Options:
  -h, --help        display help for command

Commands:
  cache [options]   inspect the local immutable Gate, Suite, and Policy cache
                    without network access
  status [options]  compare legacy local Suites with Cloud without changing
                    either side
  diff [options]    show fingerprint and DRAFT-creation changes for legacy
                    local Suites
  sync [options]    create reviewed migration results as Cloud DRAFT versions
                    only
  help [command]    display help for command
suites cache
Usage: prooflane-inspector suites cache [options]

inspect the local immutable Gate, Suite, and Policy cache without network
access

Options:
  --cache-dir <path>  owner-only immutable verified-artifact cache
  --project <id>      show one Prooflane project
  --json              print sanitized JSON
  -h, --help          display help for command
suites status
Usage: prooflane-inspector suites status [options]

compare legacy local Suites with Cloud without changing either side

Options:
  --project <id>             Prooflane project id
  --project-root <dir>       repository root containing .prooflane/suites
  --suites-dir <dir>         explicit local Suite directory
  --prooflane-token <token>
  --prooflane-url <url>
  --json                     print sanitized JSON
  -h, --help                 display help for command
suites diff
Usage: prooflane-inspector suites diff [options]

show fingerprint and DRAFT-creation changes for legacy local Suites

Options:
  --project <id>             Prooflane project id
  --project-root <dir>       repository root containing .prooflane/suites
  --suites-dir <dir>         explicit local Suite directory
  --prooflane-token <token>
  --prooflane-url <url>
  --json                     print sanitized JSON
  -h, --help                 display help for command
suites sync
Usage: prooflane-inspector suites sync [options]

create reviewed migration results as Cloud DRAFT versions only

Options:
  --yes                      confirm the displayed DRAFT-only migration plan
  --select <slug...>         synchronize only selected reviewed Suite slugs
  --project <id>             Prooflane project id
  --project-root <dir>       repository root containing .prooflane/suites
  --suites-dir <dir>         explicit local Suite directory
  --prooflane-token <token>
  --prooflane-url <url>
  --json                     print sanitized JSON
  -h, --help                 display help for command
runner
Usage: prooflane-inspector runner [options] [command]

register and operate a private Remote Runner

Options:
  -h, --help                   display help for command

Commands:
  register [options] <config>
  start [options] <config>
  help [command]               display help for command
runner register
Usage: prooflane-inspector runner register [options] <config>

Arguments:
  config                     MCP config JSON or remote URL

Options:
  --project <id>             authorized Prooflane project
  --name <name>              human-readable private Runner name
  --label <label...>         bounded routing labels
  --pool <pool>              optional private Runner pool
  --credential-file <path>   owner-only output file (default:
                             "/Users/anishohri/.prooflane/remote-runner.json")
  --prooflane-token <token>
  --prooflane-url <url>
  --url <url>
  -e, --env <pair...>
  --cwd <dir>
  -h, --help                 display help for command
runner start
Usage: prooflane-inspector runner start [options] <config>

Arguments:
  config                    MCP config JSON or remote URL

Options:
  --credential-file <path>  owner-only Runner credential (default:
                            "/Users/anishohri/.prooflane/remote-runner.json")
  --once                    claim at most one job and then exit
  --poll-ms <milliseconds>  idle polling interval (default: 3000)
  --tags-file <path>        local tagged-test registry
  --suite-file <path>       explicit migrated legacy Suite source
  --project-root <dir>      local project root
  --url <url>
  -e, --env <pair...>
  --cwd <dir>
  --json
  -h, --help                display help for command
debug
Usage: prooflane-inspector debug [options]

Options:
  -h, --help  display help for command
auth
Usage: prooflane-inspector auth [options] [command]

Options:
  -h, --help        display help for command

Commands:
  login [options]
  whoami [options]
  logout
  help [command]    display help for command
auth login
Usage: prooflane-inspector auth login [options]

Options:
  --token <token>
  --prooflane-url <url>
  -h, --help             display help for command
auth whoami
Usage: prooflane-inspector auth whoami [options]

Options:
  --prooflane-token <token>
  --prooflane-url <url>
  -h, --help                 display help for command
auth logout
Usage: prooflane-inspector auth logout [options]

Options:
  -h, --help  display help for command
project
Usage: prooflane-inspector project [options] [command]

manage deterministic Prooflane project resolution

Options:
  -h, --help      display help for command

Commands:
  bind [options]  bind the current Git remote fingerprint to an authorized
                  Prooflane project
  help [command]  display help for command
project bind
Usage: prooflane-inspector project bind [options]

bind the current Git remote fingerprint to an authorized Prooflane project

Options:
  --project <id>
  --prooflane-token <token>
  --prooflane-url <url>
  --project-root <dir>
  -h, --help                 display help for command