opendfm/xcientist/code-enablement

Code Enablement

Mission

Implement the FULL IDEA in project/ such that:

1. All-components science conditions can run with every canonical idea component enabled
2. Component-disabled science conditions can disable each idea.json component individually through the same runner interface
3. All experiments use real data from dataset_candidate/ and real API credentials

CRITICAL: Code must be SELF-CONTAINED in `project/`

• All code must be COMPLETE and INDEPENDENT in project/.
• The code must NOT depend on repos/ for core functionality.
• Everything needed to run experiments must be in project/.
• Later phases (science) will ONLY use code from project/, never from repos/.
• Repos are for REFERENCE ONLY - implement actual experiment code in project/.

Protocol

• Read agent_reports/prepare/artifacts/idea.md, agent_reports/prepare/artifacts/target_inventory.json, and validated prepare worker/reviewer reports before editing code.
• Read idea.json.components through the validated prepare handoff and preserve canonical component names.
• Bind code changes to the declared real experiment targets discovered in those reports.
• CRITICAL: Benchmark code MUST use real data files from dataset_candidate/, NOT synthetic or random data.
• CRITICAL: All experiment code MUST be under project/, NOT src/.
• CRITICAL: Each idea.json component must have a component-disable mechanism that:
• Allows the component to be disabled WITHOUT modifying other components
• Provides a clear method_context describing what the component-disabled condition does
• Is invokable by unified science conditions without changing source code between runs
• Ensure project supports: all-components runs and per-component disabled runs.
• The last step must be real final_integration_smoke run using dataset_candidate/ data.
• Remove mock fallback from formal experiment entrypoints. Mock may exist only behind explicit debug-only paths.
• API calls must read credentials from {workspace}/.env, not hardcoded keys.
• Missing models must be downloaded from HuggingFace before declaring blocker.

Report Requirements

The runtime owns reports. Do not write report files with tools. Each code step is recorded under:

• agent_reports/code/worker/<step>/attempts/NNN.json and latest.json
• agent_reports/code/review/<step>/attempts/NNN.json and latest.json
• agent_reports/code/hook/<step>/attempts/NNN.json and latest.json

Required Output

• Final worker JSON response requested by the planner; the runtime persists the worker report.
• Managed smoke evidence at agent_reports/code/artifacts/final_integration_smoke.json for final_integration_smoke
• For each idea.json component: the component-disable mechanism and method_context

Failure Conditions

• Required condition lacks runnable command
• final_integration_smoke does not use dataset_candidate/ data or real API/model
• Benchmark code uses synthetic/random data or mock vectorstores
• Experiment code placed in src/ instead of project/
• No component-disable mechanism for an idea.json component
• The all-components reference path or any component-disabled condition lacks a runnable entrypoint
• Component-disable mechanism requires modifying other components to work
• Code depends on repos/ for core functionality