The Agent Skills Directory

[PROMPT_INJECTION]: The instructions do not contain attempts to override agent behavior, bypass safety filters, or extract system prompts. The 'thinking-verb router' in SKILL.md is a benign navigational aid for the agent.
[DATA_EXFILTRATION]: There is no evidence of commands or scripts that access sensitive local files (like .env or .ssh) or exfiltrate data to remote servers. Network activity mentioned in documentation is limited to standard installation and update procedures for the repository.
[EXTERNAL_DOWNLOADS]: The README mentions a developer update script that uses git clone from github.com/ailev/FPF.git. This is a standard practice for maintainers to synchronize the skill with the upstream framework specification. No runtime remote code execution (RCE) patterns were found in the agent's instructions.
[OBFUSCATION]: No hidden or encoded content was detected. The technical and pseudo-formal language (e.g., 'U.Episteme', 'U.Holon') is consistent with the First Principles Framework theory and does not constitute obfuscation for malicious purposes.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied problem descriptions as data for reasoning.
Ingestion points: User input (problems, projects) is processed by the agent using the provided patterns (e.g., SKILL.md, sections/*).
Boundary markers: The skill uses a strict file-based navigation system (_index.md) to limit the context the agent loads at any given time.
Capability inventory: The skill only performs reasoning and navigation within its own files. It lacks file-writing or unrestricted network capabilities.
Sanitization: The agent relies on standard platform safety guardrails. The structured nature of FPF instructions tends to keep the agent focused on categorical analysis rather than open-ended execution.
[DYNAMIC_CONTEXT_INJECTION]: No !command syntax or silent shell execution triggers were found in the analyzed Markdown files.

fpf-problem-solving