php-agent-instructions

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a bundled Python script (scripts/install-agent-instructions.py) to automate the installation of instructions. The script is static, resides within the skill package, and performs standard local file operations.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to modify the agent's behavioral context by injecting instructions into project-level configuration files such as AGENTS.md or CLAUDE.md.
  • Ingestion points: The script reads existing project instructions from the local filesystem (specifically AGENTS.md or CLAUDE.md) to determine the insertion point.
  • Boundary markers: The script utilizes HTML comments (<!-- php-skills:self-review:start --> and <!-- php-skills:self-review:end -->) as delimiters. This ensures that updates are idempotent and that the skill does not inadvertently overwrite user-defined instructions outside of its designated block.
  • Capability inventory: The script performs filesystem read and write operations on the target configuration files.
  • Sanitization: The content being injected is a hardcoded block of instructions defined within the script itself, which prevents external or user-provided strings from being used to inject arbitrary prompts into the target files.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 10:34 AM
Security Audit — agent-trust-hub — php-agent-instructions