The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires terminal access to execute bundled Python scripts such as perfetto_skill.py, perfetto_query.py, and perfetto_doctor.py, along with the trace_processor_shell binary.
[EXTERNAL_DOWNLOADS]: Includes a bootstrapping script (scripts/bootstrap_trace_processor.py) designed to download and verify the checksum of the Perfetto trace_processor_shell executable if it is not found locally.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external Perfetto trace files, which represents an attack surface where adversarial data within a trace could potentially influence agent behavior. The skill mitigates this using scripts/perfetto_sql_guardrails.py to validate SQL queries before execution.
Ingestion points: Local trace files (.pftrace) provided by the user and processed via scripts/perfetto_skill.py.
Boundary markers: The operating contract emphasizes separating observations from verified root causes and requires methodology verification.
Capability inventory: Terminal execution of Python scripts and binary processors; extensive local filesystem read/write access for analysis results.
Sanitization: Employs a dedicated SQL guardrail script to check for unsafe setup and invariants before running modified or authored SQL.
[DYNAMIC_EXECUTION]: Employs a Domain Specific Language (DSL) to bind scalar inputs and JSON arrays as SQL literals, dynamically constructing and executing queries against the trace data.

perfetto-performance-analysis