The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external production telemetry data and incident reports, creating a surface for indirect prompt injection attacks.
Ingestion points: The skill reads data from the TELEMETRY_DATA parameter, as well as pre-collected telemetry files like docs/telemetry/production/latest.json (Step 1, Step 2).
Boundary markers: While the skill uses delimiter markers such as === DORA METRICS DATA START === to separate data from instructions in Step 2, it lacks explicit warnings to the LLM to ignore any instructions or overrides that might be embedded within that external data.
Capability inventory: The skill possesses several sensitive capabilities across its steps, including the ability to read and write files (Read/Write tools), spawn background sub-agents (Task tool), and execute shell commands for state persistence and knowledge sharing using the aqe and ruflo CLI tools (Step 1, Step 7, Step 8).
Sanitization: There is no evidence of content sanitization, structural validation, or safety filtering applied to the telemetry and incident data before it is passed to the core and conditional analysis agents in Parallel Batch 1 and 2.

qcsd-production-swarm