Skip to content

Experience-Driven Black-Box Cost Control for Coding Agents

1.1 KB
Raw Blame

Trace

The one place that knows each agent's session-log format. Every mechanism reads a finished session as an ordered list of (tool_name, args, result_text, success) calls through load_trajectory, whichever agent wrote it.

Module
calls.py load_trajectory (format sniffed per file), call_cwds, tool_usage, strip_rtk_prefix
catalog.py the tool list a session loaded, from whichever source the agent publishes
claude.py Claude Code transcripts: calls with the shell's directory, usage by message id, subagents, deferred tools
codex.py codex exec --json streams; a session's rollout file, converted to that stream; usage
copilot.py Copilot CLI event logs: calls, usage, prompt-turns, MCP servers
cursor.py Cursor Agent hook events, normalized to a Claude-shaped trajectory; usage is unavailable
pi.py Pi's session tree: the active branch, written as a Claude transcript; usage
jsonl.py line and record readers shared by the above

A new agent needs a reader here and nothing else: once its log reads as (name, args, result, ok) calls, every mechanism handles it.