[INFO] [<TIMESTAMP>] RepoLens run <RUN_ID> starting
[INFO] [<TIMESTAMP>] Project: <TMP_RUN_DIR>/<PROJECT_TAIL> (local/<PROJECT_TAIL>)
[INFO] [<TIMESTAMP>] Agent: codex | Mode: custom | Parallel: false
[INFO] [<TIMESTAMP>] Agent timeout: 1800s
[INFO] [<TIMESTAMP>] Agent timeout kill grace: 30s
[INFO] [<TIMESTAMP>] Lens wall-clock budget: 3600s
[INFO] [<TIMESTAMP>] Custom mode: change impact analysis (DONE streak: 1)
[INFO] [<TIMESTAMP>] Change: regression-test sentinel
[INFO] [<TIMESTAMP>] Local mode: writing local markdown files to <TMP_RUN_DIR>/issues-custom
[INFO] [<TIMESTAMP>] Resolved 1 lens(es) to run

=== Dry Run ===
Mode:         custom
Agent:        codex
Project:      <TMP_RUN_DIR>/<PROJECT_TAIL>
Rounds:      1
Lenses:       1
Output:       local markdown (<TMP_RUN_DIR>/issues-custom)

Estimated cost: ~$0.14  (lens_count=1 x depth=1 x rounds=1, lower bound — real runs typically 2-5x higher)
  model:      GPT-5 (Codex)  —  $1.25 in / $10.00 out per MTok
  repo:       ~0 source tokens  (input capped at 200k/session)
  per session: ~$0.0838  (~3000 in + 8000 out tokens)
  sessions:   1 lenses x ~1.7 iterations (streak 1 x 1.7 iter-factor) x 1 round(s)

Lenses that would run:
  security/injection

Dry run complete — no agents were executed.
