Constitutional guardrails (CONSTITUTION.md) for agent supervisor
まだ誰も着手していません。
評価
調査の方向性
Start with the affected crates listed in the issue: terraphim_agent_supervisor, terraphim_types, and terraphim_agent, then inspect the existing rust-embed and zipsign usage. Define the constitution resource, structured validation and signing boundaries, and integration tests for rejected violations. Done means the documented rules are embedded, validated before execution, logged on rejection, and covered by tests.
索引モデルが issue の本文から書いたものです。
説明
Summary
Create an immutable set of architectural principles (CONSTITUTION.md) that terraphim agents cannot violate, even when given explicit instructions to do so. Enforced by terraphim_agent_supervisor as a pre-execution validation step.
Motivation
Inspired by Ouroboros BIBLE.md pattern. The agent system crates are incomplete partly because there is no specification of what agents MUST NOT do. A constitution constrains the design space, making it easier to complete the agent layer because boundaries are clear.
Proposed Principles
- Privacy-first: No data leaves the device unless explicitly configured by the user
- Deterministic-first: Prefer Aho-Corasick automata over LLM when deterministic matching suffices
- Type-safe: No
unwrap()in library crates; all errors propagated viaResult - Compile-time checked: Prefer compile-time guarantees over runtime reflection
- Local-first: Offline operation is the default; network is optional
- Auditable: Every agent decision must be traceable to an input and a rule
- Reversible: Agent actions should be reversible where possible; destructive actions require confirmation
Implementation Plan
- Create
CONSTITUTION.mdat repo root with the principles above (editable by maintainers, versioned in git) - Embed as a compile-time resource via
rust-embed(already used interraphim_agentanddesktop) - Add
ConstitutionValidatortoterraphim_agent_supervisor:- Pre-execution hook: check proposed agent action against constitution rules
- Rules encoded as structured checks (not LLM interpretation)
- Violation = action rejected with reason, logged as event
- Cryptographically sign the constitution using existing
zipsigninfrastructure fromterraphim_update - Add integration tests: attempt constitutional violations, verify rejection
Affected Crates
terraphim_agent_supervisor(primary -- add validation)terraphim_types(addConstitutionRuletype)terraphim_agent(embed constitution resource)
Estimated Effort
~2 hours
Part of
Epic #595
- 主要言語
- Rust
- スター
- 62
- フォーク
- 5
- 平均マージ
- 2時間 27分
- マージ済み PR(30日)
- 1
コントリビューションガイド
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
terraphim/terraphim-ai のほかの issue
-
難易度 5/5 1週間以上 初心者へのやさしさ 25/100
terraphim/terraphim-ai#885 ·
-
難易度 4/5 3〜5日 初心者へのやさしさ 55/100
terraphim/terraphim-ai#871 ·
-
enhancement
難易度 5/5 1週間以上 初心者へのやさしさ 25/100
terraphim/terraphim-ai#810 · コメント 2 件 ·
-
enhancement
難易度 5/5 1週間以上 初心者へのやさしさ 35/100
terraphim/terraphim-ai#729 ·
-
enhancement
難易度 5/5 1週間以上 初心者へのやさしさ 35/100
terraphim/terraphim-ai#728 ·
terraphim/terraphim-ai の issue をすべて見る
似ている issue
-
難易度 2/5 1〜3時間 初心者へのやさしさ 75/100
-
issue
難易度 2/5 1〜3時間 初心者へのやさしさ 65/100
-
難易度 2/5 1〜3時間 初心者へのやさしさ 75/100
-
agentic-workflows
難易度 2/5 1〜3時間 初心者へのやさしさ 70/100
-
難易度 2/5 1〜3時間 初心者へのやさしさ 88/100
web-infra-dev/rspack#15847 ·