The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted text inputs (cryptic puzzles) provided by users. This creates a surface for indirect prompt injection where malicious instructions could be embedded within the puzzles to mislead the agent or exfiltrate data from its context.
Ingestion points: Puzzle text provided by the user (as noted in SKILL.md and README.md).
Boundary markers: None. The instructions do not explicitly instruct the agent to treat the input as non-executable data or to ignore embedded instructions.
Capability inventory: Execution of a local Python script (scripts/quick_decode.py) and reasoning based on extensive internal reference files.
Sanitization: No explicit sanitization or validation of the ciphertext is performed prior to processing.
[COMMAND_EXECUTION]: The skill invokes a local Python script (scripts/quick_decode.py) to perform automated decoding tasks. This is a legitimate and scoped capability used to run a script bundled with the skill. The script relies solely on Python standard libraries and treats user input as a string for transformation, which mitigates many injection risks.
[EXTERNAL_DOWNLOADS]: The reference materials (references/passcode-format.md, references/examples.md) link to external community resources such as bjres.net and github.com/ingresscodes/. These are informational links intended for manual keyword lookup or research and do not involve automated execution of remote code or unverified package installations.

ingress-decode