richfrem/agent-plugins-skills / plugins/agent-exploration-cycle/.apm/skills/running-exploration-cycle
running-exploration-cycle
VerifiedRuns the business exploration cycle end to end: starts or resumes a session (intake interview, session brief), enforces the Discovery Plan HARD-GATE, routes the four phases (Discovery Plan, Visual Blueprinting, Prototype, Handoff) to the sibling skills and keeps session state. Use when the user says "let's explore this idea", "I have an idea I want to explore", "help me plan this out", "start discovery", "resume my exploration", "where did we leave off", or starts any conversation about a need, problem or idea before building. Not for hardening an existing prototype (use reengineering-vibe-code) or for engineering build work.
references/dispatch-strategies.md
View on GitHub# Dispatch Strategies
## Contents
- [Principle](#principle) • [Strategy selection](#strategy-selection) • [Task complexity guide](#task-complexity-guide) • [Cheap Q&A pattern](#token-efficiency-cheap-sub-agents-for-qa) • [Fallback](#fallback-behavior) • [Dispatch by environment](#dispatch-reference-by-environment)
## Principle
The conductor (large context, strongest model) should do coordination and synthesis only. Delegate everything else to the cheapest runtime and model capable of the task.
**Model information is not kept here.** Runtime, model, effort and cost choices, the model catalog, and batching rules belong to the `cli-agent` skill in `agent-orchestration` (its model catalog and the dispatch guides per backend). Name no model IDs or prices in this plugin.
## Strategy selection
The expert chooses a strategy once, at session setup (Block 0 of the workflow):
| Strategy | When | How work is delegated |
|---|---|---|
| `cli-agent` (copilot, agy, claude, codex or local backends) | An external CLI is installed and authorized | Through the `cli-agent` skill; it picks the backend, model and effort and owns the batching and cost rules for that backend |
| `claude-subagents` | Claude Code only, no external CLI | The in-session Agent tool, cheap model class for simple tasks, session model for synthesis |
| `direct` | No sub-agent tooling, or everything runs in session | The current model does all of it |
Notes that outlive any catalog: per-request billing backends reward one dense request over several small ones, so batch related work before calling a premium model (details in the `cli-agent` dispatch guide for that backend). `dispatch.py` (this skill's `scripts/`) only handles the `claude`, `copilot` and `gh-copilot` backends and requires its authorization arguments (`--db-path`, `--approval-id`, `--envelope-json`, `--hmac-key-path`); `agy` goes through the `cli-agent` skill. Record the resolved strategy in the dashboard as `**Dispatch Strategy:**`.
## Task complexity guide
| Task type | Complexity | Dispatch tier |
|---|---|---|
| Single-pass document capture (requirements, stories) | Simple | Cheap |
| Business rule audit against the BRD | Simple | Cheap |
| Filling a template from provided input | Simple | Cheap |
| Re-entry scope drafting | Simple | Cheap |
| Prototype component (scoped, self-contained) | Complex | Premium |
| Handoff synthesis across several captures | Complex | Premium |
| Discovery planning session (interactive) | In-session | Conductor directly |
## Token efficiency: cheap sub-agents for Q&A
Even clarifying questions to the expert can be delegated, reserving the conductor's context for decisions and synthesis.
1. The conductor identifies a clarification need (missing information, ambiguous input).
2. It drafts the exact questions, 3 to 5 numbered.
3. A cheap worker presents them (CLI or lightweight session), collects and structures the answers, and writes `exploration/captures/clarifications-[topic].md`.
4. The conductor reads that file and continues.
**Apply to**: filling gaps in a brief, clarifying ambiguous business rules during BRD review, scoping questions during prototype observation, any Q&A expressible as numbered questions plus structured answers.
**Do not apply to**: the first framing conversation, the Discovery Planning Session (interactive, nuanced), or phase completion gates (need the conductor's judgment).
## Fallback behavior
If the chosen strategy becomes unavailable mid-session: fall back silently to `direct`, tell the expert ("The [strategy] strategy isn't available. I'll handle the rest directly."), and record `direct (fallback)` in the dashboard.
## Dispatch reference by environment
### In-session sub-agents (`claude-subagents`)
Use the Agent tool directly. Give simple tasks a cheap model class and let complex ones inherit the session model; the `cli-agent` skill lists current choices.
- Simple capture: "Read `exploration/session-brief.md`. Mode: problem-framing. Extract the problem statement, user groups, goals and scope hypotheses. Write `exploration/captures/problem-framing.md`. Mark anything not stated as `[NEEDS HUMAN INPUT]`."
- Q&A clarification: "Read `exploration/captures/brd-draft.md`. Find every `[NEEDS HUMAN INPUT]`. Present the 3 most important as plain questions, collect answers, write `exploration/captures/clarifications-brd.md`."
- Business rule audit: "Read `brd-draft.md` and `prototype-notes.md`. For each rule check confirm, contradict or uncertain. Write `exploration/captures/business-rule-audit.md`. Flag CONTRADICTED and UNCERTAIN rules."
- Handoff synthesis (complex, session model): "Read all files in `exploration/captures/`. Synthesise `exploration/handoffs/handoff-package.md` following the handoff-preparer template. Cite source files for every major claim."
### CLI backends (`cli-agent`)
Dispatch the same prompts through the `cli-agent` skill, using the prompt templates in `capturing-business-requirements/assets/prompts/` and `handing-off-to-engineering/assets/prompts/` as the persona. Pack all captures for the synthesis into one dense request. If you use `dispatch.py` directly, supply its authorization arguments:
```bash
python3 scripts/dispatch.py --agent requirements-doc-agent \
--context exploration/session-brief.md \
--instruction "Mode: problem-framing. Capture the problem statement, user groups, goals." \
--output exploration/captures/problem-framing.md \
--db-path <active_session.sqlite> --approval-id <id> --envelope-json <envelope> --hmac-key-path <key>
```
### `direct`
Run all tasks in the current session. To reduce context carry-forward: summarise each capture pass in 3 bullets before moving on, use `/compact` after Phase 1 and again after Phase 3, and pass file paths to the next phase instead of contents.
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: 322k | 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: 322k | 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: 322k | Updated | Claude CodeCodex | Verified | Install |