The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides detailed instructions for managing Fastly API tokens, including storage, rotation, and revocation. It explicitly warns agents against using commands like fastly auth show --reveal that would expose raw tokens in the conversation history, recommending instead the use of fastly auth token in shell substitutions to maintain credential privacy.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of project templates from GitHub or arbitrary URLs via the fastly compute init --from flag. It also manages the automatic download and installation of the Viceroy binary, used for local testing of edge applications, through the fastly compute serve and fastly compute install-tools commands. These downloads originate from official vendor-controlled distribution channels.
[INDIRECT_PROMPT_INJECTION]: The skill describes a workflow where the agent ingests data from external Fastly APIs and project templates (e.g., service lists, configurations, and stats) and uses this data to parameterize subsequent commands. While this creates a vulnerability surface for indirect injection if the source account or a template repository is compromised, the skill follows standard CLI operations for a developer tool and incorporates validation steps for service configurations.
[COMMAND_EXECUTION]: The skill is primarily composed of instructions for executing CLI commands to manage production infrastructure, including cache purging, WAF rule updates, and service deactivations. It mitigates risk by recommending explicit human confirmation for destructive or high-impact operations and providing pre-flight verification steps using tools like curl and openssl.

fastly-cli