The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through external error data.\n
Ingestion points: Error data is ingested from Sentry via API calls (using curl in references/ai-agent-integration.md) or via an MCP server as described in SKILL.md.\n
Boundary markers: The skill lacks delimiters or warnings to ignore embedded instructions when piping error context to the LLM (e.g., in the command cat issue.json | claude --print "Analyze this...").\n
Capability inventory: The agent is encouraged to analyze errors and "propose fixes," which implies capabilities like file system modifications and Git/PR operations.\n
Sanitization: Although PII scrubbing is implemented for data privacy, there is no sanitization to prevent the LLM from following malicious instructions contained within captured error messages.\n- [COMMAND_EXECUTION]: The skill executes shell scripts from a hardcoded path related to a dependency skill.\n
Evidence: It calls scripts located in ~/.claude/skills/sentry-observability/scripts/, including detect_sentry.sh, init_sentry.sh, and verify_setup.sh.\n- [EXTERNAL_DOWNLOADS]: The skill uses npx and curl to interact with well-known and trusted technology services.\n
Evidence: Utilizes npx to execute @anthropic/sentry-mcp, @modelcontextprotocol/server-sentry, and @sentry/wizard. It also uses curl to fetch issue details from Sentry's official API.

observability