Every dated fact stated anywhere on this site is quoted from this file, not hand-copied per page. If a number here looks old, it's a bug — file it rather than trust it.
The triage and fix-generation stages send repository file contents to the Anthropic API to do their work. Credential scrubbing applies to Patchward's own CLI output and logs — not to what is sent to Anthropic.
source: credential_proxy.py / fix_gen.py source trace, build doc D11
91.20%
source: Yehor's machine, Python 3.14.4 — same run as test-count
Runs against your real test suite in the normal (CLI-delivered) case. On the hosted webhook path, when no test runner is present, Gate 3 skips and the PR discloses this explicitly rather than proceeding silently or failing the whole fix.
source: webhook.py / pipeline.py source, §5/C2 decision (2026-08-01), build doc D7 + D8 + Decision 9
uv tool install patchward
source: callmedai.com live copy, cross-checked against real patchward --help output
CLI, run by Yehor directly against the customer's own repository
source: Triangulated from three independent Tier-0 findings: the live callmedai.com pilot CTA is a plain mailto: link with no App-install flow; the one real example PR (checkdmarc#261) was opened from a personal fork six weeks before the hosted path could push a PR with a working token (item 21 shipped 2026-08-05); and Patchward's own project memory still names "first paying Marketplace install" as an open goal as of its most recent session close. See build doc Decision 9.
>=3.12
source: pyproject.toml
Semgrep, Bandit, pip-audit, Trivy, ESLint
source: src/patchward/ pipeline source, directly inspected
565 passed, 3 skipped
source: Yehor's machine, Python 3.14.4 — real gate run, not self-reported