The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing an external dependency from a non-whitelisted source.
Evidence: skills/asc-mcp.md instructs the user to run mint install zelentsov-dev/asc-mcp@1.4.0. This downloads and installs an executable tool from a personal GitHub repository that is not part of the trusted vendors list.
[COMMAND_EXECUTION]: The skill instructs the agent to execute multiple local shell commands to perform diagnostics and management tasks.
Evidence:
skills/asc-mcp.md utilizes brew, mint, and various asc-mcp tools.
skills/testflight-triage.md and skills/app-store-diag.md use xcsym, atos, mdfind, grep, and curl to process crash logs and verify URLs.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting and processing untrusted external data.
Ingestion points: skills/testflight-triage.md (tester comments in TestFlight feedback) and skills/app-store-diag.md (App Review rejection messages from the Resolution Center).
Boundary markers: The instructions do not specify strict boundary markers (like XML tags or delimiters) when reading or analyzing this external content.
Capability inventory: The agent has the capability to execute shell commands (xcsym, atos, grep, curl) and interact with the App Store Connect API via MCP tools (e.g., reviews_create_response).
Sanitization: There is no evidence of sanitization or filtering of the content ingested from tester feedback or reviewer messages before it is processed by the agent.
[CREDENTIALS_UNSAFE]: While the skill handles sensitive API keys, it follows safe practices by instructing the user to manage them via environment variables and local configuration files.
Evidence: skills/asc-mcp.md guides the user to set ASC_KEY_ID, ASC_ISSUER_ID, and ASC_PRIVATE_KEY_PATH as environment variables for the MCP server.

axiom-shipping