Return (total_count, {domain: count}) for all tasks.

Check if a function has a for-loop that appends to a list.

Compute mean score per difficulty level from task results.

Load previously completed results from a checkpoint file.

Load curated quick-task IDs from config/quick_tasks.yaml if it exists.

Load all expert accounts from disk.

Map a human-readable domain label to the domain key.

Parse a free-text list of actions into standardized action keys.

Parse LLM output into a dict of relative_path -> content.

Parse the LLM's output into a dict of filepath -> content.

Parse JSON report with weight marker support.

Derive a stable clawId from nickname + device fingerprint. Same nickname + same device → always the same clawId, regardless of model. If the

Run LLM generation in background thread.

Load tasks and apply CLI filters.

Call via CMDOP skill system.

Run a ReAct agent loop: Think → Act → Observe → repeat.

Regenerate data/config/domains.json from TASK_ID_TO_DOMAIN.

Jaccard similarity on 3-gram word shingles.

Update existing Claw record directly (best score kept, models tracked).

Weighted overall should roughly match the reported overall.

Print comprehensive analysis.

Move selected tasks to the archive directory.

Async version — yields to event loop while waiting.

Call the LLM via the OpenAI-compatible API.

Compute progressive gain between current results and a vanilla baseline. Parameters ---------- current_results: The ``scores`` bl

Compute difficulty-weighted scores for each subject-matter domain. Parameters ---------- results: All task results (will be filte

Override rate limit for a provider.

Build a signed attestation object for a result submission.

Create a new comment on a post.

Create a new blog post.

Dedup results by profileId — keep the entry with the highest overall score.

Detect output files from instruction and verifier code.

Discover all tasks with their metadata.

Enhance a single task's verifier. Returns (added_count, summary).

Locate the vanilla baseline summary.json for a framework+model pair. Searches for ``results/{framework}-{model}-vanilla/summary.json`` relative

Fix fin-006 setup.sh that ignores $1 parameter.

Fix Python f-strings in heredocs that reference undefined WORKSPACE variable.

Fix setup.sh that copies from non-existent data/ directory.

Fix verifiers that use CLAW_WORKSPACE/os.environ instead of --workspace.

Add 'export WORKSPACE' where it's used by embedded Python but not exported.

Generate quality tier test functions.

Generate robustness test functions.

Generate strictness test functions.

Call LLM to generate all task files from a proposal. Returns a dict mapping relative file paths to their content.

Generate the next task ID for a domain (e.g., fin-001, cs-002).

Get a configuration value by key.

Get the first element from a list.

Extract the full name from a user dict.

Hash a plaintext password using bcrypt.

Initialize the database by creating all tables.

Check if the product is in stock. Returns: True if stock > 0, False otherwise.