The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data (research paper drafts in .tex or PDF format) to synthesize a health report. This creates a surface for indirect prompt injection where malicious instructions embedded in a paper could influence the agent's summary, scoring, or subsequent actions.
Ingestion points: Processes .tex source, PDF, or readable drafts; also reads .paper-memory/profile.yml and lessons.md which could be modified by other processes or skills.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when the agent reads the paper content.
Capability inventory: Writes reports to the local file system (paper-workspace/review/), executes shell scripts (reflect_log.py), and orchestrates multiple sub-skills (preflight-check, verify-citations, etc.).
Sanitization: No explicit sanitization, escaping, or validation of the paper's content is mentioned before it is processed for synthesis.
[COMMAND_EXECUTION]: The skill executes a local Python utility to update project memory with analysis results.
Evidence: append durable takeaways ... via reflect-and-improve's reflect_log.py append in SKILL.md.
Context: This is a functional component of the skill's state management, used to track recurring weaknesses and progress over time.

assess-paper