LLaDAR

English | 繁體中文 · Documentation

Generate test cases, run your Agent, and evaluate what it actually says.

LLaDAR is a command-line tool for testing Agents and chat websites. Turn your knowledge documents into questions and expected answers, collect real responses, and get a report with judgments, statistics, and the evidence behind them.

Start with a question set, or create a multi-turn situation to observe how an Agent responds as a conversation develops. Describe what you want to evaluate in plain language; use a local Skill when you need more control over the method.

What can you test?

For example, you could check whether a support Agent invents a refund policy, asks for missing information, or stays consistent when a user follows up.

An evaluator Agent judges the responses; Python computes the statistics. Saved answers and record-level evidence let you inspect the judgments and reevaluate the same responses with different criteria.

Quick start

Install LLaDAR (Python 3.11 or 3.12):

python -m pip install lladar

Configure your model provider's credentials in .env or the process environment. Your target Agent also needs its own runtime configuration. Check the installed version with lladar --version.

With your knowledge documents in ./knowledge and an Agent project in ../my-agent, run:

lladar create test-dataset --knowledge ./knowledge --output dataset.jsonl
lladar run-agent dataset.jsonl --project ../my-agent --output responses.jsonl
lladar eval responses.jsonl --output evaluation.json
lladar report evaluation.json --output report.md

This takes you from documents → test questions → actual responses → evaluation → report. Open report.md to review the results and individual test evidence. You can also use an existing question set or reevaluate saved responses.

Have a chat website instead?

Replace the run-agent command above with:

python -m playwright install chromium
lladar run-agent dataset.jsonl --page-url https://example.test/chat --output responses.jsonl

Use your own question-page URL. LLaDAR opens a browser and guides you through sign-in, discovery approval, and independent full-answer verification before running the tests. See the target selection guide for supported pages and setup details.

Decide what a good answer means

For example, evaluate whether the Agent recognizes missing information:

lladar eval responses.jsonl --criteria "Check whether the answer asks for missing information rather than inventing facts. Cite evidence from the response." --output evaluation-missing-info.json
lladar report evaluation-missing-info.json --output report-missing-info.md

For more control, provide a local SKILL.md directory with --skill to customize generation, test scheduling, evaluation, or reporting. The workflow options below explain where each method applies.

For adaptive conversations, start with the multi-turn situation example. For a complete walkthrough with custom methods, see the user guide.

Workflow options

All main workflow commands accept an optional --output directory or file path. Omitting it writes into runs/ under the current directory: a timestamped test-dataset-YYYYMMDD-HHMMSS.jsonl, situation.json, responses.jsonl, evaluation.json, or report.md. With --output ./artifacts, the same names and their sidecars go into ./artifacts/. Existing directories, trailing separators, and new paths without a suffix are directories; existing files and new paths with a suffix are filenames. For a new directory containing a dot, use ./results.v1/. Existing outputs still require --force.

Generated answers, sidecars, resume workspaces and exported adapters stay beside their main output in runs/ unless you set --output or --export-adapter. Repository tests keep temporary files in .tmp/pytest/ and their cache in .tmp/pytest-cache/; an explicit pytest --basetemp overrides that location. To preview cleanup of test/build caches, runs/, and legacy root-level responses.jsonl / response.jsonl and evaluation artifacts, run:

python scripts/cleanup_generated.py --all-generated
python scripts/cleanup_generated.py --all-generated --apply

The second command lists the candidates and asks you to type DELETE. Deleting run artifacts loses saved answers, resume checkpoints, exported adapters and captured state inside them. Tracked files, root .env / .venv, and .lladar login profiles are preserved. Existing artifacts are not moved automatically.

After each successful stage, a work summary lists the absolute output directory, each generated file and its purpose, and the next command or reading step. This summary remains visible with --no-verbose and is also saved by --log.

Generate a dataset. --output accepts an explicit .jsonl path (as below), or a directory in which the CLI creates a timestamped dataset.

lladar create test-dataset --knowledge ./knowledge --output dataset.jsonl

The bundled Akasha knowledge-point-qa skill reads source passages and generates source-grounded questions. The Agent follows the selected skill and chooses direct QA or an evidence graph for concept and controlled-variant probes; graphs are optional. Python can also produce or repair candidates. Final results are validated before publishing. Every dataset includes dataset.jsonl.generation.json; dataset.jsonl.graph.json is written only when a valid graph is used.

To let the Agent discover actual datasets in a public GitHub repository:

lladar create test-dataset --from-github https://github.com/hsinmosyi/TWBias --count 20 --output dataset.jsonl

Choose --knowledge or --from-github. GitHub mode reads repository documents, inspects CSV/JSON/JSONL or text data, and can follow documented public direct data links. It saves dataset.generation-report.md and dataset.sources.json alongside the usual generation sidecar. The report covers every output question: original samples, fixed-version links, transformations, answer origin, model review and limits. Generated criteria are identified separately from original answers or annotations; review them before using the dataset. Repository code is read for context, not executed. Private repositories, sign-in downloads, archives and binary formats are not supported. See GitHub generation for resource limits and custom Skill stages.

Graph probes observe how a target maps an inferred concept such as “meal” to source-backed peers; they do not have a single correct answer and are reported separately from correctness. Synthetic controls are test inputs, not claims about the source or people. They are off by default. Use --controlled-variant-probes in a terminal after generation lists its validated dimensions, or use --controlled-variant-topics DIMENSION_ID in a script with an ID recorded for the same corpus fingerprint. The core accepts only planner-validated dimensions, so every pair remains comparable.

This requires Akasha 1.8 or later. Use --skill DIRECTORY to override the default with one trusted local skill; there is no extra skill manifest or installation command. LLaDAR passes the one selected directory to Akasha as skills=[...]: Akasha initially exposes its metadata, then the model loads the selected SKILL.md on demand. The bundled method is not a removable user-installed skill. See skill generation for limits and Python usage.

For knowledge-optional adaptive conversations, see the situation example. Both project modes share coding-agent discovery, tools and verified repair. With max_turns=1, session calibration checks one real answer; multi-turn configurations also check memory and fresh-session isolation.

Add --adapt PATH_TO_ADAPTER.py --project PATH to reuse an existing adapter in either project mode. LLaDAR skips discovery and generation, verifies the selected protocol in project copies, and stops on failure without rewriting the supplied adapter. Keep --project: the adapter still needs the application's source and runtime.

Run the target Agent. LLaDAR inspects a copied project and creates a temporary adapter for its real public workflow:

lladar run-agent dataset.jsonl --project ../my-agent --output responses.jsonl

Which target option should you use?

• --project PATH: inspect the project's source and documentation to learn its public interface. A project alone can be enough: when startup details and the required runtime configuration are available, LLaDAR can start the service in an isolated project copy, test it, and stop only the service it started.
• --project PATH --service-url URL: inspect the same project, but use an already running test service. The URL supplies its base address, not its startup command or API contract. LLaDAR still learns the HTTP method, route, payload, authentication requirements, and response format from the project; it does not start or stop that existing service.
• --page-url URL: use a normal question webpage without providing project source. Sign in in LLaDAR's browser, handle any website dialogs, open the conversation page/window and keep the input box empty, then return to the terminal and press Enter. If several pages are open, select the conversation page by its displayed index. An Akasha Agent explores observed communication and page operations, qualifies a generated protocol/session programme, and a separate tool-free model faithfully extracts the captured answers.

HTTP, persistent SSE, native text/binary WebSocket and DOM fallback use the same discovery tools and qualification contract. The Agent investigates actual session IDs, envelopes and terminal evidence instead of assuming a website schema. Local fixtures validate these transport paths; real-site acceptance is recorded separately.

For an already running test service:

lladar run-agent dataset.jsonl --project ../my-agent --service-url http://127.0.0.1:8000 --output responses.jsonl

--service-url is optional in project mode, not an arbitrary-API discovery option.

Generating a single-question HTTP adapter with --project --service-url automatically produces a shareable API client bundle at OUTPUT.api-adapter. No export flag is needed:

lladar run-agent dataset.jsonl --project ../my-agent --service-url http://127.0.0.1:8000 --output responses.jsonl

Generation inspects the project for the HTTP contract and authentication method. A selected existing public HTTP interface must pass verification from an empty working directory using isolated Python without site packages before delivery. responses.jsonl.api-adapter contains a self-contained adapter.py, a cross-platform README.md, and a commented .env.example listing the service URL, timeout and required credential names with empty credential values. The adapter reads process environment variables; it does not load .env automatically. With --lang, non-English documentation and configuration comments are translated through akasha.ask using the adapter model; code, commands and configuration names stay unchanged. The translation is cached for resume. English needs no translation call. The Python file runs on its own: recipients need Python 3.11+, their own credentials and API network access, without the project or a LLaDAR installation. Optionally use --export-adapter api-client to customize the directory.

$env:LLADAR_SERVICE_URL = 'https://your-api.example.com'
# Set your own credential variables listed in the README or adapter.py: REQUIRED_ENV.
python responses.jsonl.api-adapter/adapter.py --message 'What can this service help with?'

--service-url URL also overrides the address. The generated adapter preserves project authentication (Bearer, API key, custom headers, cookies, or a reproducible credential exchange), reading credential values from the recipient's environment. HTTP helpers, required variable declarations, CLI and validation are folded into adapter.py: no .env, project, browser login state or run evidence is exported. LLADAR_API_TIMEOUT sets the HTTP timeout in seconds (default 60). For programmatic use, omit --message and send {"request_id":"q1","message":"question"} on stdin; stdout is JSON with the same request_id, output and observation.

The API server remains responsible for account/role/tenant/resource authorization, credential issuance, expiry and revocation. Sharing a client grants no server permissions. Missing required variables stop before requests; 401/403 are not bypassed. Redirects are rejected: specify the final API base URL. Replace localhost with a reachable server address for remote use and configure its listener/network separately; use HTTPS when sending credentials remotely. Browser login, MFA and cross-site authentication need an independently reproducible contract.

The generated README includes Python version checks, local venv setup, this adapter's required credential variable names, and launch commands for Windows PowerShell and Linux/macOS bash/zsh. Only the Python 3.11+ standard library is needed, so no requirements.txt or pip install step is included. Do not distribute the creator's .venv. Optionally add -I -S to match generation-time Python isolation. Verification covers the system running it, not every recipient OS.

Bundles currently support single-question DATASET mode with --project and --service-url. The destination must not exist; --resume accepts only an identical existing bundle. To export an older adapter, combine --adapt with --export-adapter; supplied source must satisfy the portable contract and pass verification again. Without explicit export, ordinary reused project adapters retain the existing --adapt --project flow. Generated code runs with user permissions; directory/Python isolation is not an OS sandbox. Missing startup, authentication, or request/response details require clarification; a URL alone does not fill those gaps. --page-url cannot be combined with --project or --service-url. See the target selection guide.

Use a local skill when selecting a different execution method. The bundled run-agent-stability skill runs every question three times; the run sidecars record the selected cases, repeat counts, and seed.

lladar run-agent dataset.jsonl --project ../my-agent \
  --skill ./skills/random-sample --seed 42 --output responses.jsonl

The skill receives read-only cases and schedules them through a host callback; it does not write the dataset or response files. The two packaged run skills execute their bundled, fingerprinted strategy locally, so they do not require a model call merely to calculate the schedule. Custom local run skills are still interpreted by the configured skill Agent.

When you only have an authenticated question page, provide --page-url and demonstrate one complete question and answer. You do not need to find API methods, cookies or selectors.

python -m playwright install chromium
lladar run-agent dataset.jsonl --page-url https://example.test/chat --output responses.jsonl

Run directly in an interactive terminal. Review the website, actual model destination, material scope and budgets, then type YES. In the visible browser, sign in, finish website verification, enter the service, open a new conversation, submit the full displayed demonstration question and wait for the complete answer. Return to the terminal and press Enter only when done. The program saves events, snapshots and necessary session state, then closes the recording browser.

The host builds a bounded recording index, then the material Agent selects useful events and explains why. The adapter Agent receives that material packet and can only submit complete adapt(payload) code or name a specific missing observation; it has no recording search or read tools. The host locally checks that a fresh question enters the send action, then immediately tests the candidate on the real website. A failed preflight or live trial returns the diagnostic and previous source to the adapter Agent for a new submission. Only a specific evidence gap returns work to the material Agent. Initial selection and submission each allow one model call; a missing submission stops with an error. Generation stops at the first accepted live test. The host executes the generated HTTP/SSE/WebSocket plans using its own connections. In the default independent mode, two fresh verification questions follow the accepted probe before formal dispatch. Formal execution does not depend on a visible or headless browser. Unsupported dependencies are reported instead of silently switching to DOM dispatch.

Use --demo-question TEXT to customize the demonstration; copy the full displayed question including its unique marker. --conversation-policy independent starts a fresh conversation per formal trial by default; continuous accumulates formal history in a new conversation. Demonstration and verification history are excluded. --conversation-policy demonstrated explicitly continues the demonstrated conversation; its probe and formal answers are checked for substantive, on-topic reasonableness with akasha.ask, and may depend on demonstration and probe history. This mode does not establish new-conversation support. --model selects the exploration/repair model, default gemini:gemini-3.8-flash, separately from the tested website model. Token limits follow the model profile or explicit overrides. The discovery Agents do not read future dataset questions or expected answers, or rewrite target answers.

Authorized raw recordings and scoped service state can include URLs, headers, Cookie/Authorization, bodies, client scripts, binary material and necessary storage items. Provider keys, .env, full browser profiles and unrelated website storage remain host-owned. Sign-in and MFA/CAPTCHA are manual; DOM password/OTP values are not recorded. Verbose Agent traces may contain authorized raw material; treat saved logs as private. --no-verbose retains necessary user instructions.

Every turn checks source and completion. Without an independent full-answer reference, a temporary browser permits same-turn review: MATCH if identical, MISMATCH if different, WAIT if unfinished. After three distinct trusted turns qualify the same native JSON answer path, a separate host reader can automatically check formal answers while the path and correlated natural completion remain valid. Other sources may still require manual review. Review or sign-in browsers close afterwards; the adapter independently sends and collects formal questions.

Private recordings, state, adapter, ledger and verification evidence live in OUTPUT.work/browser-discovery. Reuse with --adapter PATH/adapter/manifest.json; normally no browser opens, but fresh verification remains required. Resume with --resume OUTPUT.work; submitted or uncertain turns are recovered, never blindly resent. The old --probe-question / --browser-session-policy flags and old adapter formats are unsupported.

Set workload budgets with LLADAR_BROWSER_DISCOVERY_*, LLADAR_INDEPENDENT_*, LLADAR_PROTOCOL_* and LLADAR_MAX_*. The two Agents repair failures in alternating material-selection and code/test rounds; changed candidates need a fresh probe. Independent mode then runs two verification questions. Exhausted budgets preserve progress and stop dispatch. See the single-demonstration PRD and validation record. Local real-model success does not establish live ChatGPT, Gemini, EIP or native Streamlit acceptance.

Evaluate automatically:

lladar eval responses.jsonl --output evaluation.json

eval shows settings, per-trial progress, waiting heartbeats and results by default. Use --no-verbose to hide progress, or --log eval.log to save it.

Both evaluation modes accept --max-input-tokens N and --max-output-tokens N (positive integers). Defaults follow the selected model profile; evaluation JSON records the effective limits.

Use --criteria "Describe what to check and what counts as evidence" for ordinary evaluation requirements. Advanced users can instead supply a complete method with --skill ./skills/my-verdict; the two options are mutually exclusive. Without either, evaluation uses the default method for the record type. An explicit criteria or Skill evaluates all completed responses using that method. Eval reads the trials sidecar when present; Python calculates aggregates and correctness stability only when the plan includes boolean correct.

For situation creation, --max-input-tokens N and --max-output-tokens N independently override the authoring model's token budgets with positive integers. Omitted budgets use the selected model profile; the effective values are saved in created_with. These options apply to config creation only.

create situation --instructions TEXT (or --instructions-file PATH) plans generation, execution, stopping and default evaluation in one config. To reevaluate saved dialogue, keep the original config and add --criteria or --skill to situation eval:

lladar eval transcripts.jsonl --situation-config situation.json --criteria "Observe target terms and distinguish quotation from direct use" --output terminology-evaluation.json

The evaluation saves evaluation_settings with the effective mode, criteria text and hash, or Skill source and file evidence. Reevaluation does not modify the config or captured answers. Knowledge supplied to situation creation guides the test generator; configure the target Agent's own knowledge separately.

Render the report:

lladar report evaluation.json --output report.md

Use --skill ./skills/my-report to supply a local evidence-summary method.

eval asks one Agent to freeze a set of boolean, categorical, or numeric dimensions and judge each completed record. Counts, rates, distributions, means, medians, minima, and maxima are calculated by Python. report renders those saved facts and includes a record-level appendix.

Operation result reports

Each CLI stage command automatically attempts to save a UTF-8 Markdown operation report after success, failure, cancellation, or a request for human confirmation. This includes the five main workflows, situation modes, and five Skill-template commands. Help, version queries, and Python API calls do not create these reports.

For --output artifacts/evaluation.json, the report is artifacts/evaluation.json.operation.md. A template at lladar-skills/eval gets a sibling lladar-skills/eval.operation.md. Repeated runs use .operation-2.md, .operation-3.md, and so on, including with --force. If the output location is unresolved or unwritable, the fallback is .lladar/operation-reports/STAGE.operation.md in the original working directory.

The report follows --lang and records status, exit code, timing, permitted settings, input paths, available counts, artifact locations, sanitized problems, and next steps. It distinguishes new or modified files from existing files and includes resume guidance when the failed invocation supplies a work-session hint. Wrong answers are evaluation findings; execution or judging errors are operation issues. No extra model call is made, and the original exit code is preserved.

The report location remains visible with --no-verbose and is saved in --log when the log is active. These reports omit raw prompts, answers, URLs, credentials, and arbitrary exception messages. They complement the terminal log, generation evidence report, and evaluation analysis report. If neither location can be written, a warning is shown; forced termination cannot guarantee a report.

Language

Use --lang CODE on any workflow or Skill-template command, including --help. The default for new work is English (en), including newly generated questions and explanations. This replaces implicit source-language following.

These five languages have interface, fixed report-heading, and offline Skill-template resources. Other language codes, such as fr, are accepted for model-generated content; the interface, fixed headings, and templates use English with a short notice. Codes must contain letters and optional hyphen-separated alphanumeric subtags, such as en-US; acceptance does not guarantee that your model supports the language. Generated prose can still contain source quotes and proper names.

lladar --lang ja --help
lladar create test-dataset --knowledge knowledge.md --lang zh-TW --output dataset.jsonl
lladar run-agent dataset.jsonl --project ../my-agent --lang zh-TW --output responses.jsonl
lladar eval responses.jsonl --lang zh-TW --output evaluation.json
lladar report evaluation.json --lang ko --output report.md
lladar create report-skill --lang zh-CN

Language selection applies to progress, prompts, summaries, and newly generated LLaDAR content. Saved test questions, target Agent answers, source quotes, IDs, schema keys, code, and measurements stay unchanged. It does not ask the target Agent to switch language or add a language-scoring criterion. Situation-mode free conversation generation uses the selected language; exact probes and tokens are preserved. Website answer extraction always preserves the original answer. Technical diagnostic details and third-party traces may retain their original language.

Next-step commands include --lang. A new stage without the option uses en; resume without it restores the session's saved language. A different explicit language is rejected before further work. Sessions created without language metadata require their original version or a new work session.

Python stage APIs accept lang="ja" (the default None resolves to English for new work and the saved language on resume). Work metadata distinguishes the requested lang from the actual ui_lang; generated templates also include language.json with template_lang. Localized templates use the same tools and contracts, retain executable examples, and require no model calls or API key.

Operation logs

Verbose generation shows candidate previews as they are submitted and each final question with its expected answer. Batch-file results appear when the Agent delivers them. run-agent shows both before each trial and the actual response as soon as it arrives; missing responses are omitted. Questions are cyan, expected answers green, and actual responses magenta in color terminals. These details are flushed immediately so you can inspect them and press Ctrl+C.

Add --log PATH to any workflow command to save its terminal output to a new UTF-8 file while continuing to display it:

lladar create test-dataset --knowledge ./knowledge --output dataset.jsonl --log logs/create.log

The file includes emitted progress, Agent traces, summaries, and errors without ANSI color codes. --no-verbose suppresses detailed output in both the terminal and the log. Parent directories are created automatically; use a new filename for each run because existing logs are preserved.

With --page-url, progress shows the current discovery stage, safe tool names and ledger counts, with periodic waiting messages. Tool arguments/results and model reasoning stay out of normal logs; --no-verbose hides detailed progress.

In a Windows, Linux or macOS interactive terminal with --verbose (enabled by default), press Enter during work to see the current activity, time in that stage and time since the last progress message. This covers input checks, graph building, adapter generation/repair, model waits and target subprocess execution. Status goes to stderr and is also saved by --log. Keyboard polling pauses for human input prompts, including sign-in, choices and YES/MATCH confirmation. It is disabled with piped stdin, redirected stderr or --no-verbose. Linux and macOS retain normal canonical terminal input and Ctrl+C behavior; text typed ahead is kept for the next human prompt. Native PTY handoff tests run in the Ubuntu/macOS CI jobs.

Each LLaDAR progress message includes local time with milliseconds and a UTC offset, plus cumulative elapsed seconds ([+12.345s]). Browser waits also show the current stage's duration (stage_elapsed); the log retains the same markers.

Browser failures include safe diagnostics in verbose output and the log: the failure stage, reason, observed HTTP status or timeout limit when available, and whether a request was attempted. Later blocked trials show reason=extraction_blocked request_attempted=false blocked_by=.... Diagnostics are also saved in the trials/run sidecars; raw exceptions, URLs, headers and response bodies are excluded from these diagnostic fields.

Stream failures distinguish stream_read_failed from utf8_decode_failed. received_bytes counts bytes delivered by the browser reader after HTTP decompression, including bytes that could not be decoded. elapsed_seconds measures replay time from starting fetch to failure, excluding model extraction.

Copy-and-run walkthrough

The detailed user guide contains a PowerShell walkthrough that creates a small knowledge file, writes the complete SKILL.md contents for a randomized run, answer verdict, and evidence report, and then invokes all four commands. The only value a user must supply is the absolute path to their target Agent project. The target project and the LLaDAR model provider must be configured before run-agent can make a live call.

Each local method is just a directory containing one SKILL.md; pass that directory with --skill. There is no skill installation or management command. The walkthrough deliberately uses:

• the bundled knowledge-point-qa skill for dataset creation;
• a local run-agent-random-sample skill to select up to five cases using a supplied deterministic random helper;
• a local eval-answer-verdict skill that submits one boolean correct judgment per trial; and
• a local report-evidence-summary skill that interprets only saved facts.

See the Traditional Chinese guide for the same copy-and-run instructions in Chinese.

Record format

Every JSONL line has exactly three fields:

{"question":"What is Taiwan's capital?","expected_answer":"Taipei","actual_response":null}

question and expected_answer are non-empty strings. actual_response is a string or null. There are no schema versions and unknown fields are rejected.

Outputs

• create test-dataset: three-field JSONL with actual_response: null and <output>.generation.json; <output>.graph.json only when a valid graph is used
• run-agent: completed three-field JSONL, <output>.trials.jsonl, and <output>.run.json
• eval: evaluation plan, judgments, coverage, and aggregates in JSON
• report: Markdown tables, interpretation, limitations, and audit appendix

If browser discovery, consent, or verification stops before dataset execution, LLaDAR writes only a redacted <output>.run.json blocker record; it does not create responses or trials files.

Existing outputs are protected unless --force is specified. See the current product contract for the full behavior and safety boundaries.

Development

python -m pip install -e ".[test]"
pytest
python -m playwright install chromium
python -m pytest -m browser

Repository structure

The main source, documentation, and example paths are shown below. This is a navigation map, not an exhaustive listing; generated datasets, reports, caches, virtual environments, and build artifacts are omitted.

LLaDAR/
├── pyproject.toml              # Package metadata, dependencies, and lladar CLI entry point
├── uv.lock                     # Locked dependency versions for uv
├── README.md                   # English overview and quick start
├── README.zh-TW.md             # Traditional Chinese overview and quick start
├── qa_agent.py                 # Standalone Akasha QA script; not the lladar CLI entry point
├── src/lladar/                 # Installable Python package
│   ├── cli.py                  # CLI arguments and command dispatch
│   ├── api.py                  # Python API for creating test datasets
│   ├── skill_generation.py     # Source evidence, optional graphs, final validation, and dataset output
│   ├── skill_agent.py          # Thin adapter to Akasha's native dynamic skill loading
│   ├── method_skill.py         # Shared skill resolution and invocation for run/eval/report
│   ├── skill_assets/           # Bundled skills and supporting resources
│   │   ├── knowledge-point-qa/       # Dataset-generation method (SKILL.md)
│   │   ├── run-agent-stability/      # Repeated-run method (SKILL.md and strategy.py)
│   │   ├── run-agent-random-sample/  # Random-sampling method (SKILL.md and strategy.py)
│   │   ├── eval-answer-verdict/     # Answer-evaluation method (SKILL.md)
│   │   └── report-evidence-summary/ # Evidence-based report method (SKILL.md)
│   ├── runner.py               # Execute target agents and record responses/trials
│   ├── run_strategy.py         # Run scheduling and strategy validation
│   ├── auto_adapter.py         # Discover, verify, and replay target-project adapters
│   ├── project_profile.py      # Describe target projects from source evidence
│   ├── browser_target.py       # Discovery, qualification, ledger and formal dispatch
│   ├── browser_discovery_driver.py # Generic browser/network observations and actions
│   ├── browser_agent_loop.py   # Isolated Akasha process and owner-thread tool broker
│   ├── browser_provenance.py   # Host turn/source and full-material validation
│   ├── protocol_runtime.py    # Capability-free generated protocol programme VM
│   ├── protocol_resources.py  # Public protocol documents and pinned schema assets
│   ├── evaluation.py           # Answer judgments and deterministic aggregation
│   ├── reporting.py            # Render saved evaluation facts and report narratives
│   ├── semantic_graph.py       # Classify probe responses against source candidates
│   ├── controlled_variants.py  # Select and validate planner-declared control dimensions
│   ├── question_types.py       # Question-type and probe contracts, fingerprints, and scoring
│   ├── records.py              # Read, validate, and write dataset records
│   ├── loaders.py              # Load knowledge-source text
│   └── providers/              # Model-provider interfaces and Akasha implementation
├── tests/                      # Automated tests, browser tests, and target-project fixtures
├── scripts/                    # Live acceptance, packaging, and release verification scripts
├── example_project/            # End-to-end target projects
│   ├── diet/                   # Dietary QA agent and knowledge source
│   ├── resume_review/          # Resume-review service and console example
│   └── tainan_tutorial/        # Tainan recommendation demo, adapters, data, and tests
├── examples/support-demo-agent/ # Support-agent HTTP demo
├── akasha-agent-example/       # Small Akasha agent and hello-skill example
├── docs/                      # PRDs, design notes, verification records, and presentations
├── site/                      # English documentation website
│   └── zh-TW/                 # Traditional Chinese documentation website
├── .github/workflows/         # Release and documentation-deployment automation
└── .codex/skills/             # Development-assistant skills, separate from runtime skills

To follow dataset creation, start with cli.py → api.py → skill_generation.py; skill loading is implemented in skill_agent.py, while the method instructions live in skill_assets/*/SKILL.md. For the remaining commands, start with runner.py (run-agent), evaluation.py (eval), and reporting.py (report). To customize a method, pass a local skill directory with --skill rather than editing the bundled assets.

For target integration, also see adapter_workspace.py, project_inventory.py, graph_discovery.py, and interfaces.py for workspace inspection and discovery; target_environment.py and service_runtime.py for execution environments and managed services; and the discovery modules above, response_capture.py, and answer_extraction.py for browser operation and response extraction.

Large inputs and model limits

LLaDAR's download and sampling limits are application safeguards, rather than the language model's context limits. The current GitHub-generation budgets are:

A retained sample must also fit the selected-text and model budgets; 1,000 rows does not guarantee that a sample fits. These limits apply to GitHub generation, not to every input or browser response in the pipeline.

Model input/output token budgets are separate. Built-in generation assigns source batches derived from the configured model input budget (one quarter of the input-token budget in characters). Characters, UTF-8 bytes and tokens are different units. The model receives selected or batched evidence, rather than an entire downloaded file in one request.

Increasing --max-input-tokens or --max-output-tokens does not raise the download or sampling caps, or extend the provider's actual model capacity. Download and sampling budgets can be overridden with the LLADAR_GITHUB_* environment variables in resource_limits.py; token flags do not replace those byte/row budgets. When raising budgets, check scanning, storage, batch merging, cross-batch relationships and runtime cost.

Offline tests cover a 100,000-row CSV larger than 2 MiB, CSV fields containing newlines, limited JSONL scans, an 84,000-character local text source, and complete captures of about 1.7 MB per trial through a callable target. The callable test does not change browser capture limits. GB/TB-scale inputs have not been validated. Exceeding a budget or encountering a provider/disk failure can still stop a stage; a saved session can be resumed using the interruption instructions. Sampling and full-data processing are distinguished in the saved evidence.

See the validation record for tested scope and remaining acceptance work.

Resume interrupted work

Generation, run-agent, eval, and report save a work session next to their output, usually <output>.work. Continue explicitly with the same inputs, model, method and options:

lladar create test-dataset --knowledge ./knowledge --output dataset.jsonl --resume dataset.jsonl.work
lladar run-agent dataset.jsonl --project ../my-agent --output responses.jsonl --resume responses.jsonl.work
lladar eval responses.jsonl --output evaluation.json --resume evaluation.json.work
lladar report evaluation.json --output report.md --resume report.md.work

Use the session path printed in the work summary if it has a suffix. Python APIs accept resume=Path(...) too. --resume cannot be combined with --force. Normal runs still support --force. If a stage is interrupted, the CLI prints the working directory, input/output locations, actual session directory, and a complete quoted command to resume with the original options. A run originally authorized with --force retains that authorization in its session; the suggested resume command omits --force. Private browser/service URLs are entered locally rather than echoed into the command or log. Changed inputs, saved evidence, criteria, Skill files or model settings are rejected. Completed requests and accepted judgments are reused; a target request whose completion is unknown is recorded without automatic replay. Starting a new run creates distinct trials.

Keep the work directory until recovery and review are finished. It contains source evidence, drafts and captured responses, so apply the same access controls as to the original inputs. Credentials are not saved in the session configuration.

Built-in free-question generation uses atomic tools rather than model-written candidate files. Large CSV/JSONL sources can be streamed and reproducibly sampled; sampling limits and original row IDs are included in the generation report. A sample is not a reproduction of the original benchmark. See the implementation and validation record.

Workload limits

Model calls share the selected model profile. Browser extraction accepts both --max-input-tokens and --max-output-tokens; the default Gemini profile uses 1,048,576 input and 65,536 output tokens. Unknown models use configurable fallback request budgets of 131,072 input and 65,536 output tokens; these are defaults, not a guarantee that an unknown provider supports them. Set explicit token flags for models with a different capacity.

Memory, capture, download and waiting budgets are centralized in resource_limits.py. Every entry can be overridden before starting LLaDAR with an environment variable named LLADAR_ plus its key. Defaults include 16 MiB captured responses, 4 MiB model-input JSON, 4 MiB answer text, 32 MiB provider envelopes, 65,536 DOM nodes and 4,096 history snapshots per node. Browser demonstrations have a separate 500 MiB total recording budget (LLADAR_MAX_RECORDING_BYTES). The shared extraction deadline defaults to 3,600 seconds. Overrides become part of the resume configuration, so keep them unchanged when resuming.

$env:LLADAR_MAX_INPUT_BYTES = '8388608'
$env:LLADAR_MAX_CAPTURE_BYTES = '33554432'
$env:LLADAR_EXTRACTION_TIMEOUT_SECONDS = '7200'
lladar run-agent DATASET.jsonl --page-url URL --max-output-tokens 65536

Pagination sizes do not truncate the corpus: follow the returned continuation. Schema validation, secret detection and authorized website-request counts remain enforced independently of workload budgets.