The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data that could contain malicious instructions.
Ingestion points: Reads the latest AXIS report, including failed assistant text and goal checks (SKILL.md, Step 1).
Boundary markers: Absent; there are no instructions to ignore or delimit embedded commands within the ingested report data.
Capability inventory: The skill is capable of modifying other SKILL.md files in the repository (SKILL.md, Step 4) and executing shell commands via scripts and npx (SKILL.md, Step 5).
Sanitization: Absent; the skill does not specify any validation or filtering of the assistant text before using it to inform code patches.
[COMMAND_EXECUTION]: The skill triggers the execution of local scripts and development tools.
Evidence: Instructions in Step 5 direct the agent to run scripts/validate.sh, dora review, and npx axis run. These are standard development operations within the author's environment.
[EXTERNAL_DOWNLOADS]: The skill fetches information from external documentation.
Evidence: The skill uses Exa to fetch live product documentation from docs.scalekit.com (SKILL.md, Guardrails and Step 4). This targets the vendor's official documentation domain.

patch-from-axis