hindsight

hindsight

Regression replay across all your projects. A Claude Code plugin that loops over the test plans accumulated by iterative-tdd and re-runs them so AI-led changes get caught when they break old behavior.

Why

When you build features with iterative-tdd, every passing session writes its test plan to <project>/.tdd/regression/<slug>/. That's a free regression test — but only if something replays it. The original plugin offers a per-project replay command; hindsight is what runs them all, across every project you work on, on whatever cadence you want.

The bet: AI-led development needs aggressive regression checking, because the model that fixed the bug today doesn't remember the test plan that caught it yesterday.

What it does

~/Developer/voluntails/.tdd/regression/rbac-chunk-5/             ┐
~/Developer/voluntails/.tdd/regression/build-the-foundational/   │ 22 saved
~/Developer/va-claim-path-app/.tdd/regression/execute-phase-a/   │ regression
... 19 more ...                                                  ┘ entries

           │ /hindsight replay-all
           ▼
[hindsight discovers all entries, runs each one's saved run_command,
 compares result to the saved test plan, reports pass/fail.]

Use

As a Claude Code slash command

/hindsight-list                       # show every regression across all projects
/hindsight-list --feature auth        # scope to one feature tag
/hindsight-replay <slug>              # replay one by slug
/hindsight-replay-all                 # replay everything, exit non-zero on any failure
/hindsight-replay-all --feature auth  # scope a sweep to one feature area
/hindsight-projects                   # show which project roots hindsight will scan
/hindsight-projects add ~/Developer/new-project

Priority + feature tags (optional in replay.json)

Each replay.json may include:

{
  "slug": "...",
  "priority": "critical",     // critical | high | normal | low — default normal
  "feature": "auth",          // string OR list, e.g. ["auth", "billing"]
  "run_command": "...",
  "tests": [...]
}

/hindsight-replay-all and /hindsight-list always sort by priority (highest first) so the most important regressions get attention first. Passing --feature <tag> filters to entries that include that tag.

From the shell (CI-friendly)

python3 scripts/hindsight.py list
python3 scripts/hindsight.py replay-all
# Exit code 0 if all pass, 1 if any fail. Suitable for git hooks / CI.

How project discovery works

Hindsight looks for regression entries in this order:

1. Explicitly registered roots in ~/.config/hindsight/projects.yaml
2. Auto-scan fallback of ~/Developer/* if no config exists yet (one-time bootstrap)

You register projects with python3 scripts/hindsight.py projects add <path>.

Single flat namespace

All discovered regressions are presented as a single flat list — no project prefix in the slug. If you have collision (same slug in two projects), the first one found wins and a warning is printed; rename the colliding regression in the source project.

Where results land

Each replay writes to the source project's .tdd/regression/<slug>/runs/<timestamp>/ (same convention as iterative-tdd's replay), so the source project remains the single source of truth for run history. Hindsight aggregates the latest result from each entry in its summary output.

Status

Early. The replay path is a straight port of iterative-tdd/scripts/tdd_regression.py, extended to discover multiple project roots. Skills and slash commands are minimal wrappers. Continuous replay (watch mode, post-commit hooks) is future work.

License

MIT.