The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and follow instructions stored in local files such as CAMPAIGN.md, RUNBOOK.md, and prompts/DAILY.md. This creates a vulnerability surface where instructions from a previous session or an external actor with file access could influence the agent's behavior.
Ingestion points: File reads from the /Users/igor/work/long-term-campaigns directory tree.
Boundary markers: The instructions do not define specific delimiters or guardrails to distinguish data from instructions within these files.
Capability inventory: The skill grants the agent ability to write files, create directories, and execute shell commands for credential verification and data collection.
Sanitization: There is no evidence of content sanitization for data read from the campaign folders before it is processed by the agent.
[DYNAMIC_EXECUTION]: The skill instructs the agent to create and use scripts for "reusable deterministic collection." Generating and then executing local scripts is a dynamic execution pattern that could be abused if the generation logic is influenced by untrusted input.
[COMMAND_EXECUTION]: The skill explicitly directs the agent to execute CLI commands (e.g., gcloud) to perform "preflight" checks and authenticated requests to verify that environment access is valid.
[PERSISTENCE]: By design, this skill implements a persistence mechanism by maintaining a durable state directory (state.json, logs, and snapshots) at a hardcoded local path. This allows logic and data to survive across disposable workspace lifecycles, which is the intended functionality but inherently provides a way for instructions to persist between sessions.

long-term-campaigns