The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses dynamic execution techniques to resolve configurations and board-specific settings. It executes the output of a shell script using 'eval' in Phase 0.5 and Phase 1.1 when processing Trello boards.
Evidence: eval "$("$CLAUDE_PLUGIN_ROOT/scripts/for-each-board.sh" "[$board_json]")".
[COMMAND_EXECUTION]: The skill utilizes dynamic context injection to run a shell script when the skill is loaded. It also makes extensive use of system CLI tools to modify the environment and interact with remote services.
Evidence: !${CLAUDE_PLUGIN_ROOT}/scripts/discover-config.sh used for pre-resolved configuration.
Evidence: Use of git pull, git push, and gh CLI for repository and issue management.
[EXTERNAL_DOWNLOADS]: The skill pulls code and metadata from external sources, including GitHub repositories and Trello APIs, to facilitate the sprint process.
Evidence: git pull origin "$default_branch" and gh issue list.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external platforms and uses it to construct instructions for sub-agents.
Ingestion points: GitHub Issue bodies (gh issue list), Trello card descriptions (mcp__trello__get_cards_by_list_id), and local item files (.pm/items/).
Boundary markers: The skill looks for specific markdown headers like ## Acceptance Criteria, but does not implement strict delimiters or instructions to ignore embedded commands.
Capability inventory: The sub-agents are granted Bash, Read, Write, and Edit tools, allowing for significant filesystem and system impact if an injection is successful.
Sanitization: No explicit sanitization or escaping of the ingested issue/card content is performed before interpolation into sub-agent prompts.

sprint-dev