The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches manifest and configuration files from an external repository for data reconciliation.
Evidence: Phase 4 of the instance migration mode downloads jira-required.yaml from a GitHub repository to compare local workspace requirements with upstream templates.
Context: The download targets a well-known service domain and is used for metadata comparison rather than direct code execution.
[DATA_EXFILTRATION]: The skill accesses sensitive file paths and environment variables required to manage Jira authentication and configuration.
Sensitive paths: The skill references ~/.config/acli for machine-global session data and .env for ATLASSIAN_EMAIL and ATLASSIAN_API_TOKEN.
Context: These accesses are necessary for the skill's primary purpose of Jira administration and are accompanied by explicit instructions on secret hygiene and avoiding broad output of sensitive tokens.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external Jira instances, which could theoretically contain malicious instructions embedded in project metadata or issue fields.
Ingestion points: Data is ingested from Jira via API calls and helper scripts that read components, epics, stories, custom fields, and workflows.
Boundary markers: The skill mandates a strict 'read-first' policy and requires explicit user confirmation before any mutations are applied.
Capability inventory: The skill uses local scripts for data processing, performs network requests to Jira, and updates local configuration files in the .agents/ directory.
Sanitization: Input sanitization for Jira fields is not explicitly detailed, but the human-in-the-loop approval process for generated plans serves as a primary mitigation.

jira-administration