The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructs the agent to write tracking protocols into 'always-loaded agent instructions' such as CLAUDE.md, AGENTS.md, or .cursor/rules/ (File: SKILL.md, Step 7; METHODOLOGY.md, §5). This allows the skill to persistently modify the agent's core instructions and behavioral constraints for all future sessions within the repository.
[COMMAND_EXECUTION]: The methodology describes an automated process for uploading attachments that requires the agent to perform a direct network operation using curl to PUT raw bytes to a signed URL provided by the Linear API (File: METHODOLOGY.md, §10). This involves generating and executing a shell command with parameters derived from an external API response.
[DATA_EXFILTRATION]: The skill performs broad read operations across the repository, including README.md, CHANGELOG.md, git log, and various files in docs/ to 'backfill' project history into Linear (File: METHODOLOGY.md, §4). While the destination is a known service (Linear), this represents a large-scale transfer of repository content to an external platform.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted data from repository files (manifestos, roadmaps, logs) to derive project structure and issue descriptions.
Ingestion points: CHANGELOG.md, README.md, git log, and all files in the docs/ directory.
Boundary markers: None identified; instructions suggest the agent 'read it fully' and 'mine what already exists'.
Capability inventory: Ability to write to system instruction files (CLAUDE.md), execute network requests (curl), and create external records (Linear API).
Sanitization: Absent; instructions prioritize preserving 'useful detail' from source material rather than filtering it.
[PERSISTENCE_MECHANISMS]: The skill provides specific instructions for setting up recurring 'Automation' tasks in the Cursor editor using cron-like schedules (e.g., 0 14 * * 1) to run health checks automatically (File: AUTOMATION.md). This establishes a recurring execution pattern for the agent's skills without direct session-by-session user initiation.

linear-setup