AgentStack
SKILL verified MIT Self-run

Hyperteam

skill-samrom3-claude-hyper-plugs-hyperteam · by samrom3

Reads a session-spec, derives a task DAG, gets user approval, writes team-state.json, seeds the native task list, creates a specialist team, and monitors until the back-pressure gate passes.

No reviews yet
0 installs
8 views
0.0% view→install

Install

$ agentstack add skill-samrom3-claude-hyper-plugs-hyperteam

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

Are you the author of Hyperteam? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Hyperteam

Converts session-spec into autonomous agent team. Executes full task DAG, tracks state in plans/-team-state.json, coordinates via native task list, offers PR when gate passes.

______________________________________________________________________

Phase 0: Pre-Flight

> Prerequisites: Requires Agent Teams feature + gh CLI. Set CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1. Verify gh CLI installed and authenticated for PR creation.

Run checks in order. Stop and surface each issue as encountered.

Step 1 — Scan plans/ and select a session-spec

  1. Legacy check: List files matching *-prd.md in plans/. Any found → stop:

> Legacy -prd.md plans found: `. Rename to -session-spec.md to proceed, or create a new spec with /session-spec`. Do not silently process old-format files.

  1. List all files in plans/ matching *-session-spec.md.
  2. For each plans/-session-spec.md, determine state:
  • No plans/-team-state.jsonunstarted.
  • metadata.status = "running"in-progress.
  • metadata.status = "complete"complete.
  • Any other value → in-progress.
  1. Exclude complete specs from selection.
  2. No incomplete specs → stop:

> No incomplete session-specs found in plans/. Create a session-spec first with /session-spec.

  1. Build ordered selection list: unstarted first, then in-progress. Within each group, sort by file modification time (most recent first). Format each entry: . plans/-session-spec.md. Append for in-progress entries: ⚠ This spec may have an in-flight hyperteam run. Ensure no other session is working on it before proceeding.
  2. Single spec: Use AskUserQuestion:

> Only one incomplete session-spec found: > > plans/-session-spec.md [warning if in-progress] > > Proceed with this spec?

User confirms → select it. Otherwise stop.

  1. Multiple specs: Use AskUserQuestion:

> Multiple session-specs found. Choose one to run: > >

Wait for user's choice.

  1. Derive ` from selected filename: strip plans/ prefix and -session-spec.md` suffix.
  2. Derive ` from by stripping leading feat- prefix if present. If does not start with feat-, use as ` unchanged.

Step 2 — Checkout git branch

  1. Run git branch --show-current.
  2. Result matches `` → proceed to Step 3.
  3. Mismatch:

a. Run git branch --list . b. Branch exists locallygit checkout . c. Branch absentgit fetch origin main && git checkout -b origin/main. d. Verify git branch --show-current equals `. Mismatch → AskUserQuestion` and stop.

Step 3 — Verify symlink

  1. Run test -L plans/.
  2. Absent or not symlink → create:

`` mkdir -p plans && ln -sf ~/.claude/tasks/ plans/ ``

  1. Verify: readlink plans/ must return path ending in .claude/tasks/. Fails → AskUserQuestion and stop.

> Note: Task list scoping handled automatically by TeamCreate in Phase 2, Step 2. TeamCreate with team_name: "" creates task list at ~/.claude/tasks// and sets CLAUDE_CODE_TEAM_NAME on all teammates. No manual export of CLAUDE_CODE_TASK_LIST_ID needed.

Step 4 — Detect fresh start vs. resume

Check plans/-team-state.json:

  • Absent → Read references/phase-1-fresh-start.md and follow in full. Return here, proceed to Step 5.
  • Present → Read references/phase-1-resume.md and follow in full. Return here and proceed to Step 5 (or stop if user declines).

Step 5 — Gate discretion

After spec parsed and task DAG proposed, assess whether back-pressure gate task is needed.

Gate recommended when any of:

  • Total step count ≥ 4
  • Cross-step integration required (one step's output is another's input)
  • Output not easily human-verifiable in-session (e.g., data pipelines, infra changes)

Gate optional when ALL of:

  • Total step count ≤ 3
  • Each step has independent verify criteria (no cross-step deps)
  • Human already in verification loop (e.g., UI changes requiring manual review, or user will run claude plugin validate . themselves)
  • Unit/integration tests alone sufficient to confirm correctness

Present recommendation via AskUserQuestion:

> Gate task recommended / not recommended because [reason]. > > Proceed with gate? [Yes / No / Override]

User can override either way. If Yes (or override to Yes) → include GATE task in DAG and Phase 2 team creation. If No → omit GATE task.

Proceed to Phase 2.

______________________________________________________________________

Phase 2: Team Creation and Coordination

Step 1 — Count parallel-eligible tasks

  1. Read plans/-team-state.json.
  2. Count tasks where status: pending AND blocked_by is empty (or all listed blockers already terminal). Call this count N.
  3. Clamp to M: M = min(max(N, 1), 4).

Step 2 — Create the team

Call TeamCreate with:

  • Team name: ``
  • Teammates: 1 hyperteam-lead, M hyperteam-worker instances, 1 hyperteam-reviewer
  • Prompt includes: branch name, paths to plans/-team-state.json, plans/-progress.txt, plans/-session-spec.md

Step 3 — Seed the native task list

For every task in team-state.json with status: pending:

  1. Call TaskCreate with YAML front-matter block + full step text as description:

``` --- id: type: skills:

blocked_by:

---

```

Step 4 — Broadcast kickoff

Send broadcast SendMessage to team:

> Hyperteam ` is starting. > State file: plans/-team-state.json > Progress log: plans/-progress.txt > > All workers: claim tasks from native task list. Parse YAML front-matter in each task's description for type (FEAT or DOC) and blocked_by. Load skills: listed in task front-matter via Skill tool before beginning work. Resolve blockers via team-state.json (blocker terminal when status validated or completed). > > Reviewer: begin scanning team-state.json for completed FEAT tasks with reviewed: false` immediately.

Step 5 — Monitor

Main thread monitors run. Lead agent (dispatched in Step 2) handles coordination: review outcomes, failure resets, blocker broadcasts, GATE readiness detection.

React to events:

  • SendMessage from lead signalling GATE PASS → proceed to Phase 4.
  • SendMessage from any teammate requiring main-thread intervention → address and resume.

Main thread does not dispatch individual workers or validators. Teammates self-claim.

______________________________________________________________________

Phase 3: Back-Pressure Gate

> Runs inside reviewer agent — not main thread. Reviewer claims GATE native task when lead broadcasts GATE OPEN. See references/gate-task-template.md for full gate agent instructions.

Lead notifies main thread only after GATE passes. Proceed to Phase 4.

______________________________________________________________________

Phase 4: Completion and PR Offer

Read references/phase-4-completion.md and follow in full.

______________________________________________________________________

Phase 5: Team Cleanup

After Phase 4 completes (summary written, PR offered/created/declined), call TeamDelete for team ``. Removes all shared team resources. Must be done after Phase 4 so all teammates fully idle before cleanup.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.