Check status of all running agents. Returns list of completed ones.

Strip tool-call XML and tool-mechanism meta-chatter from a manager message so the chat reads as plain prose. Preserves real newlines/markdown.

Detect drift between the world model and reality — the failures we saw in real runs (a hypothesis left 'alive' after its experiment finished;

Validate the current scored node and run Phase 2 AutoResearch search.

Create the dsi-cluster remote workspace and record its location locally. Each dsi-slurm stage starts from a fresh runtime-owned remote root.

Convert a single DOI to BibTeX.

Extract citation-like patterns from markdown content.

Generate the comment handler prompt by combining template with idea comments. Args: idea: Full idea specification (YAML dict)

Generate the resource finder prompt by combining the template with idea specification. This is a convenience wrapper that uses PromptGenerat

Generate session instructions from template. Args: prompt: The research task prompt (from generate_research_prompt)

Get the global default paper style (used when a domain has no specific paper_style set). Returns: Paper style na

Get workspace configuration. Falls back to workspace.yaml.example template if workspace.yaml doesn't exist. This allows user

Get the workspace parent directory path. Priority (highest to lowest): 1. NEURICO_WORKSPACE environment variable (for Docker

Get workspace path for an idea if it exists. Args: idea_id: Idea identifier repo_name: Repository name (if k

Return True only when the runtime flag selects dsi-slurm compute.

Check if a stage completed successfully.

Mark this run as failed.

Move transient DSI artifacts to destination and remove the source on success.

Snapshot endpoint provenance, then execute the persisted pull manifest. The vLLM-specific bit is redacting `.neurico/modal_endpoint.json` to

Pull latest changes from remote repository. Args: repo_path: Path to local repository branch: Branch name (d

Record that an agent has completed.

Remove the recorded dsi-cluster remote workspace, then clear the record.

Move a decision from ``global`` (or any finding) onto ``finding`` — used when a decision was recorded before the finding it serves existed. Re

Run the operation and return a result.

Launch comment handler agent to make targeted improvements. Args: idea: Full idea specification with comments work_dir: Work

Run paper writing agent. The agent handles all aspects of paper generation: - Creating directory structure (paper_draft/sections/, figur

Execute research for a given idea. If GitHub integration is enabled, creates a GitHub repository, clones it, runs research t

Execute scoring/eval.py against the runner's artifact. Args: work_dir: Workspace root containing scoring/eval.py and the

Save paper results to a JSONL file, one paper per line.

Move hidden scoring files out of the workspace. Returns the sealed directory path when files were moved, otherwise None.

Return the sibling directory where sealed scoring files live. For a workspace at <workspaces>/<name>/, the sealed directory is at <works

Return the absolute path to the workspace's sentinel JSON.

Check if workspace parent directory should be auto-created. Returns: True if auto-create is enabled

Full state plus a couple of derived counts, for the dashboard.

Move hidden scoring files back to the workspace. Best-effort: logs failures but does not raise, so unseal problems do not mask the origi

Persist bootstrap baseline construction progress without marking it current-best.

Write the proposal as the first artifact of an attempt record.