The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external source (Jira) which could contain malicious instructions.
Ingestion points: The scripts/jira-view.sh script (lines 42-108) fetches issue summaries, descriptions, and comments from the Jira REST API via the go-jira CLI.
Boundary markers: The script uses text headers such as === Description === and === Comments === to delimit content, which provides some structural separation but does not prevent LLM obedience to embedded instructions.
Capability inventory: The skill possesses significant write capabilities, including issue creation (jira-create.sh), updates (jira-update.sh), transitions (jira-transition.sh), and arbitrary API access (jira-api.sh).
Sanitization: While content is transformed between Atlassian Document Format (ADF) and Markdown using scripts/_adf-to-md.js and scripts/md_to_adf.py, no security-specific sanitization is performed on the ingested text to prevent prompt injection.
[COMMAND_EXECUTION]: The skill relies on executing the go-jira binary and local scripts to perform its tasks.
Evidence: scripts/_config.sh defines a wrapper function that locates and executes the jira CLI. Several scripts (e.g., jira-create.sh, jira-update.sh, jira-view.sh) call python3 or node to run local conversion scripts (md_to_adf.py, _adf-to-md.js, _md-to-adf.js).

jira