External Campaign Triage

Use this for a campaign tracker repo (e.g. open-source-tracker) running many concurrent candidate issues/PRs across third-party repos we don't own. ai-agent-pr-metadata's picked by agent / agent:<model>-<effort>-<role> label convention only works on repos we own or collaborate on; it cannot be applied to someone else's issue or PR. This skill defines the state machine and metadata that replace it: JSON records in the tracker repo, not upstream labels.

State machine

Each candidate moves through these states, recorded in the tracker, never skipped backward:

scouted → viability_checked → claimed → implementing → in_review → merged | abandoned | stalled_awaiting_maintainer | silenced

• scouted: a candidate issue/repo identified, not yet vetted.
• viability_checked: external-pr-viability ran; only proceed past this state if it passed.
• claimed: an agent has started work. Check the tracker for an existing claimed/implementing/in_review record on this candidate before starting — the tracker file is the source of truth for "already picked" since no GitHub label exists to signal it.
• implementing: code, tests, and PR draft in progress, following subagent-pipeline roles.
• in_review: PR opened (external-pr-style for the description) and awaiting maintainer response.
• stalled_awaiting_maintainer: reached only through the Maintainer Follow-Up Guardrail below. Terminal until a maintainer responds; never re-entered from itself.
• merged / abandoned: terminal, normal outcomes.
• silenced: terminal, reached whenever a maintainer responds negatively to any contact (see Guardrails).