The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local project environment, which could potentially contain malicious instructions intended to influence the agent's planning or execution behavior.
Ingestion points: agents/analyzer.md identifies multiple files as data sources, including README.md, package.json, pubspec.yaml, build.gradle, Info.plist, AndroidManifest.xml, and existing metadata files in the metadata/ or fastlane/ directories.
Boundary markers: SKILL.md defines a mandatory 'User Approval Gate' after Phase 3 (Compliance Check). The agent is explicitly instructed not to proceed to Phase 4 (Execute) until the user provides approval for the generated plan.
Capability inventory: The skill has the capability to read project files, write new metadata files, and execute git commands as described in agents/executor.md and SKILL.md.
Sanitization: The agents/compliance-checker.md agent performs a policy scan for prohibited keywords and trademarks on the generated plan, but there is no explicit sanitization of the initial input data ingested from project files.
[COMMAND_EXECUTION]: The skill uses shell commands to interact with the project's version control system.
Evidence: SKILL.md contains a 'Repo Sync Before Edits' section that instructs the agent to run git fetch --dry-run, git status, and git pull --rebase origin "$branch" to synchronize the local repository before performing edits.

aso-marketing