shadow.example.yml

# Place at the root of your repo as `.shadow.yml`.
# Only `codebase.src_dir` is required; everything else has sensible defaults.
# This example matches a Python repo (aws-powertools/powertools-lambda-python).

codebase:
  src_dir: aws_lambda_powertools     # required: your primary source directory
  file_ext: .py                      # default "" (no extension filter)
  test_dir: tests                    # default: inferred from src_dir
  language: python                   # default "" (legacy issue-respond path only)

bot:
  name: shadow                       # used in marker comments + messages
  attribution: ""                    # optional one-line footer on posted comments, e.g. "Reviewed by Shadow" (empty = none)
  escalate_label: needs-human
  max_replies: 2                     # per-(issue, PR) cap on bot follow-up replies; 0..100, negative falls back to default
  max_runs_per_hour: 20              # per-(repo, item) hourly cap on Shadow workflow runs; 0 disables, 0..100, negative falls back to default

# Override per-stage models if you want different cost/quality trade-offs.
# Defaults below; specify only what you want to change.
# Env vars (BEDROCK_MODEL_ID, BEDROCK_REPORTER_MODEL_ID, BEDROCK_CRITIC_MODEL_ID)
# take precedence over yaml.
models:
  investigator: us.anthropic.claude-opus-4-8
  critic: us.anthropic.claude-opus-4-8
  reporter: us.anthropic.claude-haiku-4-5-20251001-v1:0   # PR-review JSON formatting
  # issue: <model>   # issue/followup answers; defaults to `reporter`. Must
  #                  # support structured output (Haiku 4.5 or Sonnet 4.6).