The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions to override the agent's perception of its security environment by claiming that network sandbox restrictions are 'stale and false' and instructing the agent to 'just make the call' despite any previous instructions or platform constraints.
[PROMPT_INJECTION]: The skill provides explicit methods to bypass security analyzers (specifically mentioned as a 'Bash secret-expansion analyzer') by using a shell expansion trick (${VAR:+x}) to hide the usage of the XAI_API_KEY in command lines and logs.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to perform network requests, build JSON payloads with jq, and write to local filesystem logs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from social media and web search results, creating a surface for indirect prompt injection attacks.
Ingestion points: Data retrieved from the X.AI API (x_search) and fallback WebSearch results.
Boundary markers: None identified; the external content is directly parsed and used to generate the final narrative map.
Capability inventory: Bash shell execution, filesystem write access (memory/logs/), and external network access.
Sanitization: None identified for the external data ingested.

narrative-tracker