All skills
simota avatar

/magi

@35ffd55
by shingo imotasimota/agent-skills85 stars
15

Deliberating decisions and founder priorities through multi-perspective, named-expert, and YC-style advisory lenses. Use for verdicts, office hours, or expert critique; not implementation.

Use this Skill: https://skilld.dev/gh/simota/agent-skills/magi

This session only. Nothing lands on disk.

referenceconclave-protocol.md

≈656 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Conclave Protocol

Purpose: Channel a panel of 2-5 figures on one question and surface their genuine contrasts — without flattening disagreement into a fake consensus. Read when: You are running the advisor conclave variant (multi-figure panel).

Principle

A conclave's value is divergence, not agreement. If three figures would genuinely clash, the clash is the deliverable. Expert Mode never resolves the clash into a verdict; transition to Magi's decision workflow only when the user actually needs to decide.

Protocol

  1. Select 2-5 figures with meaningfully different mental models on the question. Pure agreement is a weak panel; aim for productive tension. Refuse/ask per the ethics gate for each figure individually.
  2. Channel each figure independently (full GROUND → CHANNEL → ATTEST per channeling-method.md). Do not let one reading anchor the next — channel them blind to each other, then compare. (Same independence logic Magi uses for its lenses.)
  3. Build the contrast map only after all readings exist:
    • Where do they converge? (shared conclusion, possibly different reasons)
    • Where do they diverge? Name the underlying values/trade-off driving the split.
    • What question would settle it between them?
  4. Preserve disagreement. Never average two readings into a mushy middle. A 2-vs-1 split stays a split.
  5. Offer the transition. If the user must choose, pass the panel into Magi's decide or tradeoff Recipe for arbitration, or back to the user. Expert Mode advises.

Output shape

## Conclave — {Figure A}, {Figure B}, {Figure C} on {problem}

### {Figure A}  — {one-line stance}
{reading} · Attestation: A[n] I[n] S[n]

### {Figure B}  — {one-line stance}
{reading} · Attestation: A[n] I[n] S[n]

### {Figure C}  — {one-line stance}
{reading} · Attestation: A[n] I[n] S[n]

### Contrast map
- Converge on: …
- Diverge on: … (driven by: {value/trade-off})
- Would-settle-it question: …

**Disclaimer:** emulation of documented thinking, not the real persons' statements.
**Next:** decide → Magi `decide`/user · deepen one lens → `expert` · write-up → Scribe

Anti-patterns

  • Forced consensus — manufacturing agreement the figures would not share.
  • Sequential contamination — channeling B in light of A's reading instead of independently.
  • Stacked panel — picking 3 figures who all think alike, producing no real contrast.
  • Verdict creep — declaring a winner. Surface the tension; switch explicitly to Magi's decision workflow or let the user decide.

Source: SKILL.md on GitHub

1 alert13d5 checks · Risk HIGH
  • Gen Agent Trust Hub13d

    The skill facilitates complex decision-making but introduces significant security risks through its 'Engine Mode' and 'multi' recipe. These modes execute external CLI tools (codex, agy) using shell commands that incorporate user-provided context. Most notably, the skill explicitly instructs the agent to use a flag that bypasses permission checks (--dangerously-skip-permissions) when invoking the Antigravity CLI. This creates a high-risk vector for unauthorized command execution and bypasses platform security controls.

  • Socket13d

    No alerts

  • Snyk13d

    Risk: LOW · No issues

  • Runlayer6mo

    2/7 files flagged

  • ZeroLeaks5mo

    Score: 93/100 · 2 sections analyzed

Signed by skilld at 35ffd55. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 days ago.

Activeupdated 2 weeks ago

README badge

README badge for simota/agent-skills/magi