@bytesbrains/pi-ci-gate

CI observability gate for AI agents — view workflow runs, job statuses, and logs from Gitea Actions with safety controls.

Packages

Package details

extension

Install @bytesbrains/pi-ci-gate from npm and Pi will load the resources declared by the package manifest.

npm repo home report

$ pi install npm:@bytesbrains/pi-ci-gate

Package: @bytesbrains/pi-ci-gate
Version: 1.1.2
Published: May 18, 2026
Downloads: 44/mo · 44/wk
Author: nandal
License: MIT
Types: extension
Size: 42.8 KB
Dependencies: 0 dependencies · 2 peers

Pi manifest JSON

{
  "extensions": [
    "./src/index.ts"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

CI Gate for Pi

CI observability gate for AI agents — view workflow runs, job statuses, and logs from Gitea Actions with safety controls. Agents self-diagnose CI failures instead of asking humans.

Philosophy

pi-contrib-gate handles the contribution side (branch → commit → PR).
pi-review-gate handles the review side (check → approve → merge).
pi-project-gate handles the project side (issue → plan → release).
pi-ci-gate handles the CI side (observe → diagnose → re-trigger).

Install

pi install npm:@bytesbrains/pi-ci-gate

Tools

Tool	Safety	What it does
`ci_list_workflows()`	✅ Read-only	List registered workflows in the repo
`ci_list_runs(workflow?, status?, branch?, limit?)`	✅ Read-only	List workflow runs with filters
`ci_get_run(run_index)`	✅ Read-only	Get full run details (status, timing, trigger)
`ci_list_jobs(run_index)`	✅ Read-only	List jobs for a run with statuses
`ci_get_logs(run_index, job_index?)`	✅ Read-only	Get job logs (truncated to safe limits)
`ci_rerun(run_index, confirm)`	⚠️ Destructive	Re-run a failed/cancelled workflow
`ci_cancel(run_index, confirm)`	⚠️ Destructive	Cancel a running workflow

Safety Harness

The destructive tools (ci_rerun, ci_cancel) have a triple safety gate:

🛡️ Gate 1: Feature flag
   └─ .circ.yml: allowRerun: false / allowCancel: false → tool blocked entirely

🛡️ Gate 2: Explicit confirmation
   └─ Must pass confirm=true — prevents accidental triggers

🛡️ Gate 3: Rate limiting
   └─ 60s cooldown between destructive actions on the same run

Log output is also truncated: shows the head (setup) and tail (failures), keeping agents from drowning in logs while still surfacing what matters.

Configuration

Create .circ.yml:

# Max log lines returned per job (head + tail split)
maxLogLines: 200

# Enable/disable destructive tools
allowRerun: true
allowCancel: true

# Default limit for listing runs
defaultLimit: 20

Workflow

ci_list_runs(status="failure")   ← find failed runs
  │
  ▼
ci_get_run(42)                  ← inspect a specific run
  │
  ▼
ci_list_jobs(42)               ← see which jobs failed
  │
  ▼
ci_get_logs(42, job_index=1)   ← read the failure logs
  │
  ▼
[fix the code]
  │
  ▼
ci_rerun(42, confirm=true)     ← re-trigger the workflow
  │
  ▼
ci_get_run(42)                 ← verify it passed ✅

Integration

Install all four gates for full agent governance:

pi install npm:@bytesbrains/pi-contrib-gate
pi install npm:@bytesbrains/pi-review-gate
pi install npm:@bytesbrains/pi-project-gate
pi install npm:@bytesbrains/pi-ci-gate