The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute a variety of shell commands (e.g., forge deploy, forge logs, npm install) to facilitate the debugging of Atlassian Forge apps. It operates with high autonomy, having explicit instructions to run fixes immediately without requiring user permission for each step.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of untrusted data sources.
Ingestion points: The agent ingests data from forge logs (which can include arbitrary output from the application under test) and local source code files (manifest.yml, package.json, and JavaScript/TypeScript files).
Boundary markers: The instructions do not define boundary markers or include warnings for the agent to ignore instructions that might be embedded within the logs or source code.
Capability inventory: The agent has extensive capabilities, including the ability to write to the file system (to inject debug logs or apply fixes), install software packages via npm, and deploy applications to the Forge platform.
Sanitization: There is no evidence of sanitization, filtering, or validation performed on the ingested data before the agent uses it to make decisions or execute commands.
[EXTERNAL_DOWNLOADS]: The skill manages and installs the @forge/cli tool and other project dependencies through npm. These actions target official Atlassian packages and well-known public registries, which are consistent with the skill's purpose and the developer's infrastructure.

forge-debugger