hassan-mohiddin/freeflow / .deprecated/delegation-harness/skills/delegation-harness
delegation-harness
VerifiedUse when coordinating Freeflow Pi/cmux pane delegation, orchestrator/planning-parent/execution-parent workflows, task packets, child results, context locality, capability reroutes, multi-agent execution, work packages, worktrees, routing fresh-reviewer/subagent requests, or when the user asks to spawn/manage visible pane agents.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 12.0 KB |
| references/context-locality.md | markdown | 1.5 KB |
| references/execution-and-integration.md | markdown | 3.9 KB |
| references/reviewer-profiles.md | markdown | 3.8 KB |
| references/roles-and-contracts.md | markdown | 4.1 KB |
| references/task-packets-and-results.md | markdown | 5.1 KB |
| references/tool-policy.md | markdown | 1.7 KB |
references/roles-and-contracts.md
View on GitHub# Roles And Contracts
Use roles to preserve ownership boundaries. Parent and child describe context topology, not competence or seniority. Use parent reports to move context between phases without continuously supervising responsible workers.
## Orchestrator
Owns global continuity, user-facing routing, final closeout, final commit/push decision, and completion claim.
The orchestrator watches parents by default, not every leaf. It may inspect any descendant, but should normally instruct through the direct parent.
## Planning-Parent
Owns deep planning conversation and artifact creation.
It may do basic scouting inline and launch researchers for broad, deep, or specialized evidence. It runs the standing independent review of a consequential spec/plan package before implementation; other artifact reviews require Workflow's scoped authorization.
Planning-parent writes specs and plans because it has the planning context. It self-reviews its artifacts and may read `review-artifact` for richer lenses without creating independence.
Outputs `PLANNING_REPORT`:
```text
status
spec/plan/review paths
settled decisions
open questions
execution autonomy expectation
expected user checkpoints
execution guidance
risks
evidence pointers
```
## Execution-Parent
Owns plan-guided execution.
It builds the live execution map, assigns work packages, launches workers or an integrator when useful, dispatches approved plan-selected phase-exit reviews, adjudicates formal findings, handles planned intermediate commits, and reports back. At final assurance it freezes one state and launches a fresh verifier plus a different fresh reviewer in parallel. Neither consumes the other's output; the parent collects both before adjudication. Other reviewers or independent verifiers require Workflow authorization.
For broad or multi-slice implementation, it assigns a worker to the implementation stream instead of self-implementing slice by slice. It decides from the whole execution package, not from whether the next slice looks small.
Execution-parent may edit for coordination, reporting, or mechanical integration. If it edits product/runtime files, it must state why that edit is not worker-owned.
Outputs `EXECUTION_REPORT`:
```text
status
packages completed/not completed
commits
reviews
checks
files changed
plan deviations
stop conditions hit
open questions
risks
recommended closeout
evidence pointers
```
## Leaf Children
- **Researcher**: deep/broad evidence gathering. No mutation.
- **Worker**: responsible engineer for the assigned package in its checkout/worktree. It implements, self-verifies, self-reviews, may read review/verify skills for richer inline guidance, corrects local reversible mistakes, and learns across sequential slices; do not replace it per slice by default.
- **Reviewer**: evidence-backed independent second opinion at a selected consequential boundary. It does not supervise the worker continuously, own source truth, fix findings, or count as the verifier.
- **Verifier**: distinct fresh context that runs the finalized allowed checks and reports factual evidence without editing or reviewing design. The standing final verifier runs once; extra independent runs require user authorization.
- **Integrator**: merge/apply worker outputs and resolve integration issues within scope.
Leaf output is a role-native structured result, report, or blocker. Workers report changed files/checks/findings. Reviewers report blocking/non-blocking/questions without fixing. Verifiers report check evidence and unverified claims. Leaf children do not own product decisions, source-truth changes, final closeout, or push.
## User Conversation
The user can always talk to the orchestrator. During planning or execution, the user may talk to the active parent. The user is the accountable owner and collaborator: they own intent and consequential decisions, while live evidence owns factual behavior.
Agents should correct unsupported factual or technical claims with evidence rather than agree performatively. Direct user intervention in a leaf is allowed for debugging but should be logged and surfaced to the direct parent.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| 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: 472k | Updated | Claude CodeCodex | Verified | Install |
| 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: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| 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 available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |