The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple external sources including conversation history, repository diffs, commits, issues, pull requests, and CI evidence to reconstruct a workstream narrative.
Ingestion points: SKILL.md (Workflow Step 1 and 2) identifies that the agent must resolve the workstream boundary from commits, issues, PRs, and reviews.
Boundary markers: The instructions provide guidelines to define boundaries and avoid inventing narratives, but do not provide explicit technical delimiters or sanitization rules for the ingested data.
Capability inventory: The skill utilizes Python 3 for telemetry analysis and delegates mutations like documentation edits and issue creation (create-issue).
Sanitization: There is no evidence of specific escaping or filtering of data ingested from the workstream evidence.
[DYNAMIC_EXECUTION]: The skill invokes Python 3 to perform data analysis on telemetry files.
Evidence: SKILL.md states that 'Timing ledger analysis uses Python 3 when available telemetry calls for it.' This implies runtime generation or execution of Python scripts based on telemetry content.
[COMMAND_EXECUTION]: The skill instructs the agent to execute specific commands located in a file outside the skill's own directory.
Evidence: references/timing-analysis.md directs the agent to 'run the bundled one-shot validate and summarize commands from event-ledger.md' located at a relative path (../../milestone-rush/references/event-ledger.md).

run-retro