Skip to content

heymegabyte/agent-skills / skills/paperclip

paperclip

Verified

Paperclip AI Control Plane (swarml.ink) governance, sub-task decomposition, Reviewers/Approvers/Monitor/Watchdog fields, Parent/Blocking/Blocked-by graph relations, Decisions contract (clarifying questions, recommendations & scope expansions only), and trunk-based main PR auto-merge + production deployment.

Paperclip (swarml.ink) Operating & Governance Skill

Canonical operating instructions for interacting with the Paperclip AI Control Plane (https://swarml.ink, loopback http://127.0.0.1:3100 on ubuntu.megabyte.space, Company ID 72408921-0094-4ffb-b4f4-f24db9026191).


1. Use Sub-Tasks (parent_id / parentId) Whenever Appropriate

Whenever an issue represents an epic, multi-package initiative, cross-repository triad sync (agent-skills + install.doctor + theme.tools), or multi-stage workflow:

  1. Create Child Sub-Tasks (parentId / parent_id):
    • Decompose the parent issue into focused child sub-tasks via POST /api/companies/{companyId}/issues with "parentId": "{parentIssueId}" (or parent_id in PostgreSQL).
    • Assign each child sub-task to the best-fit specialist agent from the 29-agent paramilitary roster (Full-Stack-Engineer, Backend-Engineer, Database-Engineer, DevSecOps, UI-Designer, QA-Verification, AppSec, A11y-Auditor, SEO-Auditor, Release-Manager, etc.).
  2. Automatic Parent Wakeup:
    • When all direct child sub-tasks reach a terminal state (done / cancelled), Paperclip automatically fires PAPERCLIP_WAKE_REASON=issue_children_completed to wake the parent issue's assignee for final integration and closure.

2. Leverage "Parent", "Blocking", and "Blocked by" Relations Whenever Appropriate

Structure is expressed with Parent (parent_id), while execution ordering is enforced with Blocking (blocks) and Blocked by (blockedBy / blockedByIssueIds, stored in issue_relations with type = 'blocks'):

  1. Parent (parentId / parent_id):
    • Links a sub-task to its parent epic/initiative so work products, rollup progress, and issue_children_completed wakes stay unified.
  2. Blocked by (blockedByIssueIds / blockedBy) & Blocking (blocks):
    • Whenever Issue B depends on Issue A finishing first (e.g., schema/contract lock before implementation, shared @theme.tools/* library before consumer site integration, or implementation before QA/release), set "blockedByIssueIds": ["{issueAId}"] on Issue B.
    • Setting blockedByIssueIds on Issue B automatically populates both Blocked by on Issue B and Blocking on Issue A.
    • When all blockedBy issues reach done, Paperclip automatically wakes the dependent issue's assignee with PAPERCLIP_WAKE_REASON=issue_blockers_resolved.

3. Leverage Reviewers, Approvers, Monitor, and Watchdog Fields on Every Issue

Every active issue in Paperclip (todo, in_progress, in_review, blocked) must leverage all four governance and liveness fields:

  1. Reviewers (execution_policy.stages stage type: "review" + review_policy: "not_creator"):
    • Populate Stage 1 (type: "review", approvalsNeeded: 1) with independent specialist reviewers excluding the issue's assignee:
      • QA-Verification (f299bd09-55b8-46b2-bcf0-0cd832a20591)
      • AppSec (59f8d581-168e-4013-92a2-53f6288d3bcd)
      • A11y-Auditor (f75dd55c-e2c0-491f-981b-85e3947f5960)
    • Set review_policy = 'not_creator' so no agent can self-review its own work.
  2. Approvers (execution_policy.stages stage type: "approval"):
    • Populate Stage 2 (type: "approval", approvalsNeeded: 1) with architectural/release approvers excluding the issue's assignee:
      • Architect (a6e94825-1999-4b5f-ad32-a94d5354db0d)
      • Release-Manager (ef69998f-9aae-4118-a28e-dae33194ba31)
      • CEO (48b44790-8e8a-44ca-8989-e871d41b51e5)
  3. Monitor (execution_policy.monitor + monitor_next_check_at, monitor_notes, monitor_scheduled_by):
    • On every in_progress or in_review agent-owned issue (assignee_agent_id IS NOT NULL, assignee_user_id IS NULL), schedule an active issue monitor (monitor_next_check_at, monitor_notes, monitor_scheduled_by = 'assignee', plus execution_policy.monitor and execution_state.monitor).
    • Paperclip's tickDueIssueMonitors scheduler automatically wakes the assignee (PAPERCLIP_WAKE_REASON=issue_monitor_due) when monitor_next_check_at arrives so no in-progress PR or deployment ever stalls.
  4. Watchdog (issue_watchdogs table: watchdog_agent_id, instructions, status = 'active'):
    • Attach an active watchdog entry in issue_watchdogs (SRE 4ac23ce6-e37f-473d-834e-48b6bf824f3f or Swarm-Dispatcher 60c08948-5f59-4ffe-bde3-19dc26074404) on every open issue to audit heartbeat liveness, recover stalled runs, and enforce trunk-based main convergence.

4. "Decisions" (/MEG/decisions) — Strictly User Questions, Recommendations & Scope Expansions

In Paperclip (https://swarml.ink/MEG/decisions), Decisions are used exclusively to ask the user (kcWlxnR12wGUuh8CNKjy3zHU061tHaI2) high-signal questions that:

  1. Clarify ambiguous instructions or product/architectural trade-offs, OR
  2. Provide the user with concrete recommendations and scope expansions for approval.

Rules for Creating Decisions (ask_user_questions)

  • Use POST /api/issues/{issueId}/interactions with:
    • "kind": "ask_user_questions"
    • "resolverPolicy": "human_only"
    • "continuationPolicy": "wake_assignee"
    • "addresseeUserId": "kcWlxnR12wGUuh8CNKjy3zHU061tHaI2"
    • Clear title, summary, and payload.questionSet / payload.questions where the best option is first and prefixed with (Recommended) plus a custom free-text option.
  • What Must NEVER Appear in Decisions:
    • Never create Decisions or interactions for non-questions, routine status updates, bureaucratic role-to-role handoffs, asking permission to merge a green PR into main, or internal blocker bookkeeping.
    • Database triggers (trg_paperclip_decisions_questions_only and trg_paperclip_sync_interaction_decision_retention) and scripts/paperclip-watchdog.ts automatically archive any non-question noise so /MEG/decisions only ever displays genuine questions that make sense for the user to answer.

5. Trunk-Based main (shared_workspace) PR Creation, Auto-Merge & Production Deployment

  1. Work on main (execution_workspace_preference = 'shared_workspace').
  2. Pull latest origin/main, implement and test all OpenSpec tasks, open a GitHub PR targeting main (gh pr create --base main), and immediately merge the PR into main (gh pr merge --merge --delete-branch).
  3. Deploy merged main changes to production (wrangler deploy + CDN cache purge, npm publish to registry.megabyte.space, Coolify deployment, or systemd reload on ubuntu.megabyte.space), verify live health, and post the merged PR URL, main commit SHA, and production proof on the Paperclip issue before marking done.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall