femto-retry-probe

/femto:retry-probe

Used after a hook block (KC mastery below threshold) when the engineer wants to re-engage with new probe questions on the same session.

Pre-checks (in this order)

1. Resolve the session_id.
2. Call femto_get_session_status({ session_id }). Refuse with a clear message if:
   • phase is graded or failed — session is terminal; start a new one.
   • grader_status.subagent_dispatched == true && !grader_status.grader_md_present — grader is in flight; wait for it.

Reset steps

1. Back up probe-log.md (preserve audit trail):

   sessdir=".femto/session-<session_id>"
   ts=$(date -u +%Y%m%dT%H%M%SZ)
   if [ -f "$sessdir/probe-log.md" ]; then
     cp "$sessdir/probe-log.md" "$sessdir/probe-log.md.$ts.bak"
     : > "$sessdir/probe-log.md"
   fi
   

2. Clear probe state in state.json by editing coverage and probe_log to empty objects/arrays, and ensure phase: probe_active. Use a minimal jq rewrite:

   jq '.coverage = {} | .probe_log = [] | .phase = "probe_active" | del(.failure_reason) | del(.probe_last_turn_at)' \
     "$sessdir/state.json" > "$sessdir/state.json.tmp" && mv "$sessdir/state.json.tmp" "$sessdir/state.json"
   

3. Call femto_advance_probe({ session_id }) to confirm reset. The first call should return a current_kc and coverage showing zero turns for every required KC.

4. Tell the engineer: probe state cleared; previous transcript saved at <sessdir>/probe-log.md.<timestamp>.bak. Continue probing as normal.

Constraints

• Always back up before truncating. Audit trail is load-bearing per Gate 1 transparency.
• Do not reset reads_status. Reads are still valid; only probe state needs clearing.
• Do not call femto_request_grading here. That happens after the new probe terminates.