Skip to content

lawzava/megapowers / plugins/megapowers/skills/orchestrating

orchestrating

Verified

Use when two or more independent lanes can run in parallel, an output-only lane protects lead context from raw payload, or a handoff or durable run needs coordination. Covers fan out, spawn agents, split the work, delegate lanes, review while building, and independent files or services to inspect at once.

FileKindSize
SKILL.mdskill md3.3 KB
assets/agent-capabilities.mdmarkdown3.1 KB
references/fences.mdmarkdown1.5 KB
references/native-dispatch.mdmarkdown5.2 KB
references/review-contract.mdmarkdown1.3 KB

assets/agent-capabilities.md

View on GitHub
# Agent capabilities

Personal model preferences for lead and delegated agents. Copy this file to
`~/.config/megapowers/agent-capabilities.md` and replace every `<placeholder>`.
User instructions, repository instructions, available tools, and permissions
take precedence. Apart from the review rule below, this file grants no access,
write, or effect authority.

```yaml
version: 2
refreshed_at: <YYYY-MM-DD>
expires_at: <YYYY-MM-DD>   # about a month out; an expired file is ignored

policy:
  capability_floor: strong             # weakest profile intelligence a ranked lane accepts
  optimize_after_floor: [speed, cost]  # tie-break order among eligible bindings
  escalate_on: [oracle_failure, high_risk, unresolved_ambiguity]

profiles:
  fast-read: { roles: [explorer], intelligence: strong, write: false }
  balanced-build: { roles: [writer], intelligence: strong, write: true }
  challenger: { roles: [challenger], intelligence: frontier }
  deep-consult: { roles: [consultant], intelligence: frontier, write: false }
  independent-review: { roles: [reviewer], write: false }

lead_preferences:
  <harness>: { model: <model-id>, effort: high }

bindings:
  <harness>:
    fast-read:
      model: <model-id>
      effort: low
      agent_type: <native-agent-type>
      family: <vendor-family>
      access: native
      rankable: true
    balanced-build:
      model: <model-id>
      effort: high
      agent_type: <native-agent-type>
      family: <vendor-family>
      access: native
      rankable: true
    independent-review:
      agent_type: <native-reviewer-role>  # role configured by the user, not shipped
      model: <model-id>
      effort: high
      family: <vendor-family>  # e.g. anthropic, openai, or google
      access: native
      status: operator-selected
      rankable: false

fallbacks:
  fast-read: { provider: <harness>, model: <model-id>, effort: medium, write: false }
```

## Rules

- Define what counts as independent and which reviewer to use in your global
  CLAUDE.md or AGENTS.md. A different vendor is preferred but optional unless
  your instructions require it. Bind a model and route that are available.
- An `independent-review` binding is your standing approval to send review
  material to that reviewer within the stated scope. Check briefs and accessible
  files for secrets and disclosure scope before sending anything off-machine.
  Use existing native subagents or native review. The
  plugin supplies no review launcher or automated disclosure checks.
- An unavailable binding is reported, not replaced. Use a `fallbacks` entry
  only when access, disclosure, and permissions for it already exist.
- A failed task oracle calls for diagnosis, not a provider switch.
- `intelligence`, `speed`, and `cost` are relative operator judgments, not
  measurements. Record the evidence and date for any benchmark you rely on
  in a note below the block.
- Effort names do not mean the same amount of thinking across models. Set
  effort from the vendor's default for each new model and your own runs at
  several levels, not from the previous model's value.
- Keep private access data, command lines, private paths, and routing secrets out of
  this file.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 323kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 323kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 323kUpdated Claude CodeCodexVerifiedInstall