Skip to content

v0.1.0

Choose a tag to compare

View all tags
@zufardhiyaulhaq zufardhiyaulhaq released this 30 Oct 04:26
· 8 commits to master since this release
v0.1.0
7a55676

pergent v0.1.0

First release. pergent runs agentic merge request reviews in CI by driving opencode with your own review skills, and posts one comment with a verdict per skill.

ghcr.io/gopaytech/pergent:v0.1.0

Reviews

  • Skills are Markdown files with paths/exclude scope; load them one by one with --skill or all at once with --skills-dir.
  • Batched, parallel runs. Changed files are split into batches that fit the model's context, and every skill and batch runs in parallel under one --timeout budget and one --concurrency limit.
  • Structured findings with a blocking or suggestion severity. Blocking findings are re-checked by a separate verification run before they count.
  • Verdict per skill: no_blocking_issues, has_blocking_issues or failed. A failed run is reported in the comment and never passes as clean.
  • Summaries and checklists show what each review checked and which files it covered.
  • Default skips for deleted, binary, lockfile, vendored and generated files.
  • Previous review: --previous-review feeds the last comment's findings into the next run.

Platforms

  • GitHub and GitLab. The comment is updated in place instead of duplicated, and stays within each platform's size limit.

Agent mode

  • pergent agent reviews --base..--head in a checkout you prepare and delivers a versioned JSON result (schema v1) to a file, stdout or an HTTP endpoint. It holds no platform token. See docs/agent-result-schema.md.

Security

  • opencode runs as a read-only agent: no edits, shell, web access or .env reads, and the reviewed repo's own opencode config is ignored.
  • Tokens and keys are redacted from the comment and from logs, including opencode's stderr.
Assets 2
Loading