Projects + PR Monitoring

IntelligenceX includes a GitHub-native maintainer control plane for PR and issue backlog monitoring using the intelligencex todo ... command family.

Important disclaimer

These commands are assistive. They are not an autonomous production decision system.

• Treat category, tags, IX Suggested Decision, duplicate clusters, and scope classification as recommendations.
• Keep final decisions human-owned (Maintainer Decision, merge, close, defer).
• Do not wire this into unattended production mutation flows without explicit human review gates.

What this gives you

• Bot-review checklist sync into TODO.md (sync-bot-feedback).
• Backlog indexing and duplicate clustering across open PRs/issues (build-triage-index).
• Scope alignment checks against VISION.md (vision-check).
• Issue applicability review for stale/no-longer-applicable infra blockers (issue-review).
• Observe-mode PR babysitter snapshots and action recommendations (pr-watch).
• Issue applicability proposed actions (close, keep-open, needs-human-review) with confidence scoring and safety signals.
• GitHub Project field sync for triage at scale (project-init, project-sync, project-bootstrap).
• Maintainer-assist view checklist and apply plan generation (project-view-checklist, project-view-apply).
• Signal quality grading (high/medium/low) to separate strong recommendations from weak-context items.
• Operational PR signals (PR Size, PR Churn Risk, PR Merge Readiness, PR Freshness, PR Check Health, PR Review Latency, PR Merge Conflict Risk) for faster triage.

Recommended end-to-end flow

1. Pull explicit bot checklist items into TODO.md.
2. Build triage index artifacts.
3. Run issue applicability review.
4. Run vision alignment check.
5. Bootstrap or initialize GitHub Project.
6. Sync triage + issue-review + vision into project fields.
7. Generate project view checklist and apply plan for maintainers.

End-to-end example

# 1) Sync explicit bot checklist items to TODO.md
intelligencex todo sync-bot-feedback --repo EvotecIT/IntelligenceX

# 2) Build triage index artifacts
intelligencex todo build-triage-index --repo EvotecIT/IntelligenceX

# 3) Review infra blocker issue applicability (dry-run, with streak state)
intelligencex todo issue-review --repo EvotecIT/IntelligenceX --proposal-only --min-consecutive-candidates 2 --min-auto-close-confidence 80 --state-path artifacts/triage/ix-issue-review-state.json

# 4) Check backlog against VISION.md
intelligencex todo vision-check --repo EvotecIT/IntelligenceX --vision VISION.md

# 5) Bootstrap project + workflow + vision scaffold
intelligencex todo project-bootstrap --repo EvotecIT/IntelligenceX --owner EvotecIT

# 6) Sync triage + issue-review + vision into project fields
intelligencex todo project-sync --config artifacts/triage/ix-project-config.json --issue-review artifacts/triage/ix-issue-review.json --max-items 500

# 7) Generate a maintainer view checklist
intelligencex todo project-view-checklist --config artifacts/triage/ix-project-config.json --create-issue

Command quick map

Automation workflow

• Workflow: .github/workflows/issue-review.yml
• Schedule: nightly dry-run (no auto-close by default)
• Manual auto-close: use workflow_dispatch with:
  • apply_close=true
  • confirm_apply_close=CLOSE_ISSUES
  • min_auto_close_confidence (default 80)
  • optional label policy (allow_labels, deny_labels)

Observe-mode babysitter automation:

• Workflow: .github/workflows/ix-pr-babysit-monitor.yml
• Triggers:
  • event-driven on PR lifecycle and submitted/edited reviews (pull_request, pull_request_review)
  • hourly observe-mode safety-net sweep
• Engine command: intelligencex todo pr-watch-monitor
• Behavior:
  • event-driven runs target the triggering PR automatically,
  • PR-triggered monitor runs are forced onto GitHub-hosted runners (security hardening),
  • pull_request_review_comment is intentionally excluded to avoid duplicate paired monitor runs from the same review action,
• workflow_dispatch supports manual targeted runs with optional pr override and policy inputs (max_prs, max_flaky_retries, include_drafts, approved_bots)
• retry_failure_policy controls whether retry suggestions use legacy any mode or the opt-in smarter non-actionable-only mode
• Outputs: per-PR snapshots + rollup summary + audit log (ix-pr-watch-audit.jsonl) in artifacts/pr-watch/
• Monitor and nightly rollups now include failed workflow-run totals plus actionable / operational / mixed / unknown evidence counts

Guarded retry assist automation:

• Workflow: .github/workflows/ix-pr-babysit-assist-retry.yml
• Trigger: manual workflow_dispatch only (single PR target)
• Engine command: intelligencex todo pr-watch-assist-retry
• Safety: requires explicit confirmation token RETRY_CHECKS
• Scope: retries failed checks only when pr-watch plans an eligible retry_failed_checks action (dedupe + cooldown aware)
• retry_failure_policy=non-actionable-only makes retry planning suppress likely code/test breakages while still allowing operational or unknown failures
• Audit: emits execution outcomes (success/skipped/failed) into artifacts/pr-watch/ix-pr-watch-audit.jsonl

Nightly consolidation automation:

• Workflow: .github/workflows/ix-pr-babysit-nightly-consolidation.yml
• Schedule: daily consolidation sweep (plus workflow_dispatch and reusable workflow_call)
• Engine command: intelligencex todo pr-watch-consolidate
• Inputs: max_prs, stale_days, include_drafts, approved_bots, source
• Inputs also include optional retry_failure_policy (any by default, non-actionable-only opt-in)
• Optional tracker issue inputs: publish_tracking_issue, tracker_issue_title, tracker_issue_labels
• Optional governance labeling input: apply_governance_signal_label (default false)
• Weekly governance scheduled runs can also opt into tracker governance labeling via repo variable IX_PR_WATCH_APPLY_GOVERNANCE_SIGNAL_LABEL=true; manual runs continue to use the explicit apply_governance_signal_label input
• Outputs:
  • rollup JSON: artifacts/pr-watch/ix-pr-watch-nightly-rollup.json
  • markdown summary: artifacts/pr-watch/ix-pr-watch-nightly-summary.md
  • metrics JSON: artifacts/pr-watch/ix-pr-watch-nightly-metrics.json
  • metrics history JSON: artifacts/pr-watch/ix-pr-watch-nightly-metrics-history.json
  • tracker issue body markdown: artifacts/pr-watch/ix-pr-watch-nightly-tracker.md
• Nightly rollups and tracker summaries also surface failed workflow-run kind breakdowns so maintainers can judge whether failures look infra-like or code-like before adjusting retry policy
• Nightly metrics also emit conservative retry-policy guidance based on consecutive failure-profile trends, for example suggesting non-actionable-only only after repeated operational/unknown-dominant runs
• When that guidance becomes stable enough to recommend a change, weekly/nightly tracker output treats it as a governance signal, but it remains advisory only; workflow inputs are not changed automatically
• Nightly metrics now also expose that recommendation as a machine-readable governanceSignals block so downstream governance/reporting flows can sort on policy-review suggestions without scraping markdown
• Nightly and weekly markdown summaries now also include a compact Governance: line near the top so maintainers can spot active policy-review recommendations without opening tracker details
• If apply_governance_signal_label=true, the tracker issue will also manage the repo-facing label ix/retry-policy-review-suggested based on the current governance signal; this remains opt-in and is removed again when the signal clears
• triage-project-sync.yml can also opt into syncing that live governance signal onto PR project items via the managed label ix/pr-watch:policy-review-suggested, using workflow_dispatch input apply_pr_watch_governance_labels or repo variable IX_TRIAGE_APPLY_PR_WATCH_GOVERNANCE_LABELS; this path is also off by default
• triage-project-sync.yml can independently opt into syncing optional project fields PR Governance Signal and PR Governance Summary from the same live tracker state by using workflow_dispatch input apply_pr_watch_governance_fields or repo variable IX_TRIAGE_APPLY_PR_WATCH_GOVERNANCE_FIELDS; this path is also off by default
• triage-project-sync.yml can also independently opt into the optional Governance Review project view profile by using workflow_dispatch input include_pr_watch_governance_views or repo variable IX_TRIAGE_INCLUDE_PR_WATCH_GOVERNANCE_VIEWS; this keeps the default view set unchanged for teams that are not using the governance fields
• Generated project configs now persist that governance intent in features.prWatchGovernance; project-view checklist/apply uses it for optional governance view coverage, and todo project-sync uses it for governance label/field defaults when a config file is present
• todo project-sync also supports --no-apply-pr-watch-governance-labels and --no-apply-pr-watch-governance-fields, so config-backed defaults remain overridable per run
• Consolidation buckets include:
  • stale infra-like blockers,
  • review-required/stuck PRs,
  • retry-budget-exhausted PRs,
  • no-progress PRs grouped by age/churn class.
• Tracker issue behavior:
  • source-scoped upsert using marker <!-- intelligencex:pr-watch-rollup-tracker:<source> -->,
  • auto-enabled for scheduled/reusable runs, opt-in for manual dispatch.

Weekly governance automation:

• Workflow: .github/workflows/ix-pr-babysit-weekly-governance.yml
• Schedule: weekly consolidation sweep (Monday)
• Implementation: calls the nightly consolidation workflow via workflow_call with weekly profile defaults (max_prs=300, stale_days=14, source=weekly-governance)
• Manual workflow_dispatch also exposes optional overrides for retry_failure_policy, tracker publishing, governance-signal labeling, tracker title/labels, and the weekly scan window so teams can opt in deliberately without changing the scheduled defaults
• Outputs: same rollup/summary/metrics schema as nightly consolidation for comparable trend analysis
• Tracker issue: enabled by default (publish_tracking_issue=true) for weekly governance runs
• Stable retry-policy recommendations can keep the weekly governance tracker issue open even when blocker buckets are otherwise quiet, so maintainers can opt in deliberately

Cadence matrix

This cadence keeps mutation guarded and targeted:

• observe runs are event-driven plus scheduled (hourly/daily/weekly),
• retry assist remains explicit/manual per PR (ix-pr-babysit-assist-retry.yml),
• no autonomous merge or policy mutation is introduced.

Issue-review confidence signals

issue-review now emits a proposed action and confidence score for each issue:

• proposedAction: close, keep-open, needs-human-review, ignore
• actionConfidence: 0-100 with level hints (high/medium/low)
• confidenceSignals: explainable signal list (for example stale bucket, recent activity, linked PR age, reopened count)
• project-sync maps these into project fields: Issue Review Action, Issue Review Action Confidence

Safety behavior:

• Auto-close requires both policy eligibility and confidence threshold (--min-auto-close-confidence).
• Recently active issues and reopened issues are downgraded to needs-human-review.
• Use --proposal-only for calibration runs where any close operation must be blocked.

Permissions and safety

• Project setup and sync requires GitHub project scope (read:project also required for sync reads).
• Issue-posting helpers require issue write permission.
• Use --dry-run on sync commands before enabling mutating operations.
• Treat low-signal items (Signal Quality = low, ix/signal:low) as context-gathering tasks, not decision-ready recommendations.

Related docs

• Project Bootstrap and Sync
• Project Views and Operations
• Reviewer Overview