AGENT SKILL

dream

Deep creative exploration engine. Competing perspectives, stress-tested by a 4-persona council, scored by integrity. The approach that survives attack wins — not the one that sounds best.

License 许可
Apache-2.0
最近更新
2026-11-14
Stars
★ 1.6k
Forks
491

01 前言 Front Matter

name
dream
kernel
{ "kind": "workflow", "version": 1, "confirmation": "none", "side_effects": "writes_meta" }
description
Deep creative exploration engine. Competing perspectives, stress-tested by a 4-persona council, scored by integrity. The approach that survives attack wins — not the one that sounds best.
allowed-tools
Agent, Bash, Read, Write, Grep, Glob, WebSearch
user-invocable
true

03 SKILL.md 正文

<skill id="dream">

<purpose> Expand the solution space BEFORE committing.

Not "give me 3 options" — structured divergence: competing value systems + adversarial stress test + integrity scoring.

The winning approach is the one that SURVIVES attack, not the one that sounds best.

Use before any non-trivial decision. Use when the obvious answer feels too easy. </purpose>

<on_start> agentdb read-start # prior dreams + learnings seed the perspectives; don't re-explore killed approaches </on_start>

<skill_load> always: skills/quality/SKILL.md, skills/architecture/SKILL.md on_domain: api: skills/architecture/reference/api.md, skills/architecture/reference/backend.md frontend: skills/frontend/SKILL.md backend: skills/architecture/reference/backend.md security: skills/tearitapart/reference/security.md </skill_load>

<!-- ============================================ --> <!-- THE DREAM CYCLE --> <!-- ============================================ -->

<phase id="0_ground" name="Ground in Reality"> Before dreaming, understand what exists:

  1. Glob/Grep affected areas in the codebase
  2. Check _meta/research/ for prior work
  3. Check agentdb for related learnings/failures
  4. Map existing patterns, conventions, constraints

Dreams that ignore the codebase are fantasies, not proposals. </phase>

<phase id="1_diverge" name="Generate Competing Perspectives">

<perspective id="minimalist" voice="terse, provocative, reductive"> Goal: the SMALLEST possible solution. Question the premise itself. - Can we delete our way to the answer? - Does an existing tool/library already do this? - What if we just... don't build this? - What's the 20-line version?

Target: 90% code reduction. Must reference actual files that could be deleted.
Format: 3-8 lines. Effort estimate. Coverage percentage.

</perspective>

<perspective id="maximalist" voice="expansive, visionary, system-thinking"> Goal: the version you'd be PROUD of in 6 months. - What does the ideal architecture look like? - What does this unlock beyond the immediate need? - What edge cases should be handled from day 1?

Target: complete solution. Must sketch actual architecture, not hand-wave.
Format: full description with component diagram. Effort estimate.

</perspective>

<perspective id="pragmatist" voice="balanced, explicit about tradeoffs, deadline-aware"> Goal: the 80/20 point. Ship this week. - What's the minimum that solves the real problem? - What can we defer without paying interest? - What's the upgrade path when we need more?

Target: 80% solution with clear upgrade path.
Must explicitly state what's deferred and the cost of deferral.
Format: concrete plan. Effort estimate. Tradeoff table.

</perspective>

Tier 1: generate all 3 inline. Tier 2+: spawn dreamer agent for codebase-grounded perspectives. </phase>

<phase id="2_stress_test" name="4-Persona Council — Find What Breaks"> For EACH perspective, run through the council. Not voting — adversarial probing.

<council> <persona id="architect" concern="fragility, scale, tech debt"> Probes structural integrity. Coupling, single points of failure, migration nightmares. Breaks things by asking "what happens when..." </persona>

<persona id="user" concern="usability, complexity, does it solve MY problem?">
  Cuts through elegance. "Does it work for the person using this every day?"
</persona>

<persona id="adversary" concern="what breaks, worst case, what was missed">
  Pure attack mode. Edge cases, race conditions, security holes, wrong assumptions.
  If they can't find a flaw, the approach is strong.
</persona>

<persona id="operator" concern="can we ship it, can we maintain it, blast radius">
  Operational reality. Deployment, monitoring, rollback, on-call burden.
  Beautiful code that's hell to operate fails.
</persona>

</council>

Each persona: 2-3 lines per perspective. Specific concerns, not essays.

<ask_user> Use AskUserQuestion when: perspectives generated, before running council Ask: "Three perspectives ready. Which resonates, or run all through stress test?" Options: stress test all, lean toward {minimalist|maximalist|pragmatist}, rethink framing </ask_user> </phase>

<phase id="3_measure" name="Integrity Scoring"> For each perspective, score integrity based on council feedback:

  • How many council members raised critical (not fixable) concerns?
  • Did the perspective already account for the concerns?
  • Are the flaws structural or cosmetic?

= 0.8: ANTIFRAGILE — stronger because of the attacks. = 0.6: VIABLE — survives with minor fixes. < 0.6: SHATTERED — fundamental flaws. Don't pursue.

Rank surviving perspectives by score. If ALL shatter: the problem needs reframing (thermal shock). </phase>

<phase id="4_present" name="Present Results"> <output_format>

Dream: {topic}

Context

{codebase state, constraints, existing patterns}

Perspectives (ranked by integrity)

{emoji} {name} — integrity: {score}

{perspective content} Effort: {estimate} Council verdict: {1-line summary per persona} Survived because: {why it's robust} — {perspective name}

{repeat for each surviving perspective}

Shattered

{any that didn't survive, with reason}


Recommendation

{highest integrity + why. hybrid options if scores are close.}

Next: /kernel:forge {approach} or /kernel:ingest for guided execution. </output_format>

<ask_user> Use AskUserQuestion when: results presented with ranked perspectives Ask: "Proceed with {winner}, hybrid approach, or rethink the problem?" Options: proceed with winner, hybrid of top 2, rethink </ask_user> </phase>

<phase id="thermal_shock" name="All Perspectives Shattered" trigger="all_shatter"> Every approach failed the stress test. The problem needs reframing.

  1. Record why each shattered (agentdb learn failure)
  2. Ask: is the problem statement wrong? Solving the right thing?
  3. Generate 1-2 reframings of the original problem
  4. Return to diverge phase with reframed problem (max 1 reframe)
  5. If still shatters: STOP. "This needs human decomposition." </phase>

<github_integration> If gh authenticated and profile is github-oss or github-production: Post dream to GitHub Discussions (Decisions category). Otherwise: Write to _meta/dreams/{topic}.md only. </github_integration>

<telemetry> agentdb emit command "dream" "" '{"topic":"X","perspectives":3,"survived":N,"chosen":"pragmatist","integrity":0.85}' </telemetry>

</skill>

04 同仓库更多 Skills

hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-review-with-annotations

Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-find-and-synthesize-context

Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-update-docs

Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-records

Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-interactive-html

Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-collaborate-threads

Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.

★ 1.6k