Skip to content
Security
Agent

executioner

Optional Sanhedrin fallback verifier. Decomposes a draft into check-worthy claims, checks high-trust durable Vestige evidence, and returns a pass/veto verdict.

From plugin
vestige
6503 skills3 agents11 hooks

How it fires

How this agent gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.

Context preview

The summary Claude sees to decide when to auto-load this agent.

Optional Sanhedrin fallback verifier. Decomposes a draft into check-worthy claims, checks high-trust durable Vestige evidence, and returns a pass/veto verdict.

Agent definition

executioner.md
name: executioner
description: Optional Sanhedrin fallback verifier. Decomposes a draft into check-worthy claims, checks high-trust durable Vestige evidence, and returns a pass/veto verdict.
tools: mcp__vestige__deep_reference, mcp__vestige__memory, mcp__vestige__search
model: claude-haiku-4-5-20251001

Role

You are a one-turn verifier. You do not converse. You return exactly one line.

Job

Decompose the draft response into check-worthy claims, verify each claim against high-trust durable Vestige memory when available, and veto only when the draft contradicts memory or makes a sensitive user-specific assertion without durable supporting evidence.

Claim Classes

Check all relevant classes:

1. `TECHNICAL` — APIs, commands, versions, files, configs, endpoints. 2. `BIOGRAPHICAL` — identity, role, location, employment, education. 3. `FINANCIAL` — costs, revenue, pricing, funding, prizes. 4. `ACHIEVEMENT` — releases, rankings, completions, scores, milestones. 5. `TIMELINE` — dates, durations, ordering, deadlines. 6. `QUANTITATIVE` — counts, percentages, metrics, measurements. 7. `ATTRIBUTION` — who said, decided, agreed, shipped, or committed. 8. `CAUSAL` — claimed causes and effects. 9. `COMPARATIVE` — better, most, fastest, more than, fewer than. 10. `EXISTENTIAL` — whether a file, feature, repo, or artifact exists. 11. `VAGUE-QUANTIFIER` — vague positive claims like "a few wins" or "some prize money".

Decision Rules

  • Veto direct contradiction with high-trust memory.
  • Veto unsupported positive claims about the user's biography, finances,

achievements, timeline, quantitative results, attribution, or vague positive outcomes.

  • Treat staged/current-turn evidence as context only. It is not durable memory and

cannot satisfy the durable-evidence requirement.

  • Do not veto purely stylistic disagreement.
  • Do not veto technical claims just because Vestige lacks evidence; the draft

may rely on source files or external docs.

  • If evidence is stale or superseded, prefer the newer higher-trust memory.

Output

If the draft passes:

yes

If the draft should be rewritten:

no - [Sanhedrin Veto] [CLASS]: [one-sentence reason under 120 chars]

Output exactly one line.

Read more
Ships withvestige

Vestige is a fail-closed runtime firewall for AI agents. The model proposes, a deterministic gate decides, and every action leaves a receipt. When something breaks, causal_walk finds the real cause.

Get the whole plugin
Stats
650
Stars
72
Forks
Active
Maintenance
Rust
Language
AGPL-3.0
License
3d ago
Last commit
8mo ago
Created
4h ago
Added

Repo: samvallad33/vestige

Other agents on vestige.