The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/generate_report.py dynamically loads and executes Python modules from the local filesystem using importlib.util to maintain parity with core campaign tools.
Modules loaded: run_job_campaign.py, campaign_guard.py, and plot_progress.py.
Path resolution: Paths are computed relative to the skill's installation directory, targeting the nvflare-autofl/scripts/ neighbor directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted campaign data which could potentially contain malicious instructions intended to influence the agent reading the generated reports.
Ingestion points: Reads results.tsv, autofl.yaml, and candidate manifests to synthesize campaign findings.
Boundary markers: The script uses standard markdown table formatting and basic cell escaping, but lacks explicit delimiters to warn the agent about potentially untrusted content in the resulting report.
Capability inventory: The skill is restricted to file system operations (reading logs and writing reports) and loading local scripts within the campaign workspace.
Sanitization: Basic escaping is performed for markdown table structure (e.g., escaping pipe characters).

nvflare-autofl-report