The Agent Skills Directory

[COMMAND_EXECUTION]: The skill heavily utilizes shell commands and bash scripts to perform its auditing functions. This includes orchestrator scripts like scripts/check-conformance.sh which calls specialized scripts for architecture, coding standards, and documentation. Additionally, checkpoints.yaml contains numerous mechanical checks implemented as shell command strings.
[DYNAMIC_EXECUTION]: The skill performs token-level static analysis by executing inline PHP code via the php -r command in checkpoints.yaml (see entries TC-120, TC-156, and TC-157). This is used to process PHP tokens more reliably than regex, specifically for detecting prohibited shell functions or validating translation keys.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests untrusted code and documentation from the TYPO3 extensions it audits. Malicious instructions placed in the comments, metadata, or templates of a target extension could attempt to manipulate the agent's audit findings, particularly during the LLM-based review phases defined in the llm_reviews section of checkpoints.yaml (TC-60 through TC-100h).
Ingestion points: The agent reads the contents of the Classes/, Configuration/, Resources/, and Documentation/ directories of the extension being audited.
Boundary markers: The LLM review prompts provided in checkpoints.yaml do not currently include explicit boundary markers or instructions to ignore embedded commands within the ingested data.
Capability inventory: The agent has the capability to execute shell scripts and report findings back to the user based on its analysis of external data.
Sanitization: Mechanical checks use grep, find, and tokenization which are robust, but the LLM review phase relies on the underlying model's instructions.

typo3-conformance