The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, specifically analysis findings and the full text of a deliverable document, and provides this content to an LLM for semantic verification.
Ingestion points: Data is ingested via a JSON task file containing the findings array and deliverable string (referenced in SKILL.md and scripts/fidelity-check.mjs).
Boundary markers: There are no explicit boundary markers or instructions defined to prevent the LLM from obeying instructions embedded within the processed deliverable text.
Capability inventory: The skill utilizes the Bash tool to run a local verification script and performs LLM inference for spot-checking critical findings.
Sanitization: The input text is normalized for white space and casing in the mechanical script, but no sanitization or escaping is applied to the content before it is processed by the LLM.
[COMMAND_EXECUTION]: The skill invokes a local script to perform deterministic analysis on the input data.
Evidence: The workflow requires executing node scripts/fidelity-check.mjs <plik-zadania.json> via the Bash tool.
Context: The script is a local, zero-dependency Node.js utility provided with the skill. It reads the local input file and performs keyword matching and stemming to identify missing findings. It does not perform network operations or access sensitive system paths.

deliverable-fidelity-pl