The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project files to automate releases and compliance reporting, creating an attack surface where malicious content in project files could influence agent behavior.
Ingestion points: Processes project manifest files (AndroidManifest.xml), dependency trees (releaseRuntimeClasspath), and store metadata fields across several files including SKILL.md and references/degraded-mode.md.
Boundary markers: Implements a 'GATE' protocol requiring explicit human approval for all irreversible or externally visible actions (detailed in references/idempotency.md).
Capability inventory: Possesses capabilities to build and sign artifacts, upload to stores, and manage environment secrets via the GitHub CLI.
Sanitization: Instructions focus on generating answer sheets for human review (e.g., Data Safety forms) rather than programmatic sanitization of ingested content.
[COMMAND_EXECUTION]: The skill orchestrates numerous system commands and internal scripts to perform release tasks.
Evidence: Uses gh (GitHub CLI), fastlane, xcodebuild, gradlew, keytool, openssl, and altool for build and secret management.
Evidence: Invokes internal plugin scripts such as python3 skills/play-store-publish/references/upload_aab.py for credential verification and uploads.
[CREDENTIALS_UNSAFE]: The skill provides instructions for managing and verifying highly sensitive release credentials.
Evidence: Guides the agent to locate, base64 encode, and verify Google Play Service Account JSON keys, Apple Distribution Certificates (.p12), and App Store Connect API keys (.p8) stored in sensitive local directories like ~/.appstoreconnect/private_keys/.
Note: The skill proactively advises against committing secrets to version control and promotes the use of secret managers like GitHub Secrets, aligning with security best practices.

release-automation