v0.2.0

Minor Changes

• 27fffee: Let a launcher choose the session ID before starting an instrumented program, through the sessionId layer option or the EFFECT_INSPECT_SESSION_ID environment variable; a random UUID remains the default. The ID survives reconnects, while a different run reusing an ID the collector already holds is refused and counted as a conflict instead of being merged into the original trace. Collision refusal needs the updated collector; older collectors still merge runs that reuse an ID.
• 1a9a75e: The collector now answers read-only trace queries over HTTP on its port (POST /api/v1/query, GET /api/v1/export?sessionId=ID): session discovery, a run summary, filtered spans, one span with its ancestry, children and events, and correlated logs. Answers are bounded JSON that name the exact session and time reference, report recorded loss or unknown completeness, and refuse a session whose ID another run reused. Saved traces answer the same queries offline, and an exported trace keeps the collector's loss counters in an optional header field that older builds ignore.
• a2aba7c: New query commands for coding agents: summary, spans, span, logs, export and sessions. Launch an instrumented program with EFFECT_INSPECT_SESSION_ID=my-run-001, then query exactly that run with --session my-run-001, or a saved trace with --file PATH and no collector. Each command prints one bounded JSON document on stdout (--json for one compact line), a diagnostic on stderr on failure, and a distinct exit code per outcome. export writes a lossless .eitrace without overwriting an existing file unless --force. Root and per-command --help document the workflow, flags, JSON fields, errors and exit codes. The interactive --wizard global flag is no longer offered.