Evidence

Evidence (prints)

English · Português (Brasil)

Evidence is disabled by default. When enabled, the app records prints of the whole window (without masks) on the baseline and on every detected change:

%TEMP%\screen_watch\captures\<target>\<YYYYMMDD-HHMMSS-mmm>_<baseline|change>.png

The <target> is the selection file name (e.g.: painel). Action runs record with the _action suffix (or _action-<step>).

How to enable

From the window (simplest): the Record captures (evidence) checkbox persists in state.json["evidence_enabled"] and applies without editing the YAML — including with config v1 (targets:). It takes precedence over the evidence: section of the YAML.

From the config (v2):

evidence:
  enabled: true
  dir: null              # null = %TEMP%/screen_watch/captures
  keep_per_target: 50    # keeps the N most recent ones per target
  max_total_mb: 200      # total cap (all targets)
  on_baseline: true      # record a print of the baseline
  on_change: true        # record a print of every change
  per_step: false        # in the actions: one print per step (in addition to the run print)

Retention prunes the folder after each write: by count (keep_per_target) and by size (max_total_mb).

Where they live

• Default: %TEMP%/screen_watch/captures (outside the repository — the prints stay only on your machine).
• With evidence.dir configured, the prints go there.
• python -m screen_watch show-paths shows the effective folder on the captures: line (with a note when there is an override).

The Captures button in the window opens the effective folder in the file manager, creating it if it does not exist yet; if the loop prints are off, it warns in the log. Opening a folder/file always goes through platform/shell.py::open_path (os.startfile on Windows, open/xdg-open on the others); if there is no association/utility, the GUI just logs "open manually: ".

Generate prints right away (without waiting for a real event)

These flows always record, even with loop evidence off:

python -m screen_watch test-evidence --selection painel    # baseline+change; prints the paths
python -m screen_watch test-action --selection painel --armed   # print of the action run

The window's Run action (3s) button also records a print when running. In the actions, with per_step: true, one print per step is recorded and the paths appear in the audit (logs/actions.jsonl).

Failures

Failures when writing (permission, disk, window off-screen) only generate log.warning; monitoring continues.

Related

• Configuration — the evidence: section and state.json
• GUI and tray — checkbox and print buttons
• doc/00 §11.5 — design decisions