The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a workflow that generates and executes Python scripts at runtime. It uses the scaffold_signal_engine tool to create a signal_engine.py script based on templates and agent-extracted logic, which is then executed using the backtest tool. This pattern of script generation and execution presents a risk if the generation process is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from academic papers and reports via the read_document tool. The extracted information (formulas, entry/exit rules) is used to populate executable code templates. This ingestion of external, attacker-controllable data into a code generation pipeline constitutes a vulnerability surface.
Ingestion points: read_document(paper_path) in SKILL.md (Phase 1).
Boundary markers: None identified; the skill does not instruct the agent to ignore or delimit instructions found within the processed documents.
Capability inventory: bash, backtest, scaffold_signal_engine, and sdm_register in SKILL.md and referenced files.
Sanitization: The skill uses ast.parse via a bash command for syntax validation, which identifies malformed Python but does not check for malicious logic or side effects.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill uses the bash tool in Phase 3 to execute a Python syntax check on generated code. Additionally, it recommends manual installation of the rapidocr_onnxruntime package via pip in documentation if OCR capabilities are missing. The primary risk lies in the execution of agent-generated code via the backtest tool.
[PERSISTENCE_MECHANISMS]: The skill documentation in references/scheduled_decay_scan.md describes the configuration of a ScheduledResearchExecutor. This mechanism allows for the persistent scheduling of agent tasks (decay scans) that survive server restarts and execute prompts periodically.

strategy-dev-manager