Test the floor against a real shell instead of hand-written shell rules
メンテナーはふだん 1 日以内に返信
まだ誰も着手していません。
評価
- 難易度
- 4/5
- 見積もり時間
- 3〜5日
- 初心者へのやさしさ
- 62/100
- issue の種類
- 機能追加
- 明瞭さ
- 明確に書かれている
- 活発さ
- 活発
- 技術スタック
- bash, typescript
調査の方向性
Start with floor.ts and evaluateFloor, then inspect how this repository runs tests or scripts. Add a bash-backed fixture with the named command stubs and capture stdout, stderr, file writes, and piped sinks across the listed command shapes. Done means the oracle agrees with evaluateFloor for observable exposures, with file sinks checked against policy, and the test is isolated from the default bun test path if needed.
索引モデルが issue の本文から書いたものです。
説明
floor.ts decides whether a secret reaches a sink by reading the command with hand-written rules. Across four review rounds on #85, every blocking finding was the same mistake: those rules were written from memory of what a shell does, and each fix was right for the case in front of it and wrong for the next spelling.
- Round 1: the host tokenizer strips quotes, so a quoted capture read as a print.
- Round 2: the fd digit came off its redirect, so
2>/dev/nullread as the allowed sink. - Round 3:
nohup,commandandbuiltinwere listed as assignment prefixes. They are not, and the shell's error message prints the secret. - A background review: a backwards scan let a literal
envin argument position restore assignment position, soecho env TOKEN=$(op read …)read as a capture.
What eventually worked was asking the shell. Build it into the repo as a fixture.
Shape
For each command shape, run it in a real bash with stub secret sources on PATH, then compare what the shell did to what evaluateFloor answers.
- Stubs:
security(prints the secret only with-w),op,pass, and apbcopythat appends to a capture file, all in a temp dir prepended toPATH. - Oracle: the secret is exposed when it appears in stdout, stderr, a file the command wrote, or a sink process it was piped into. Capture all four, not just stdout: a first draft that watched stdout alone reported four false mismatches.
- Assertion:
evaluateFloor({command}).asks === exposed, for every shape where the oracle can observe. A file sink asks by policy even when the oracle cannot see the write, so those rows are asserted against the policy instead.
A working version ran 24 shapes (captures quoted, unquoted and backticked; every redirect spelling including &>, &>>, >&, 2>, 1>&2; the wrapper prefixes; quoted read commands) and agreed with the shell on 23, the last being an oracle limitation rather than a floor bug.
Why it is worth the fixture
It converts the recurring class from "a reviewer notices a spelling" into a failing test, and it is the only source consulted so far that has not been wrong. It also gives #90 (time misclassified as an exec wrapper) a natural home: add the shape, watch it fail, move the word.
Cost: it shells out, so it belongs behind its own script or a tagged test rather than in the default bun test path if CI images cannot be trusted to have bash.
- 主要言語
- TypeScript
- スター
- 0
- フォーク
- 1
- 平均マージ
- 3時間 35分
- マージ済み PR(30日)
- 50
環境構築
このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。
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。
STRML/omp-classifier のほかの issue
-
Decide whether a coordinator may lift a headless worker's refusal (the trust boundary #68 defers)オープンenhancement ready-for-human
難易度 5/5 1週間以上 初心者へのやさしさ 25/100
STRML/omp-classifier#142 ·
メンテナーはふだん 1 日以内に返信
-
enhancement ready-for-human
難易度 4/5 3〜5日 初心者へのやさしさ 45/100
STRML/omp-classifier#116 · コメント 8 件 ·
メンテナーはふだん 1 日以内に返信
-
enhancement ready-for-human
難易度 5/5 1週間以上 初心者へのやさしさ 35/100
STRML/omp-classifier#13 · コメント 6 件 ·
メンテナーはふだん 1 日以内に返信
STRML/omp-classifier 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 72/100
NousResearch/hermes-agent#136483 ·
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 86/100
メンテナーはふだん 1 日以内に返信
-
factory-active factory-automatic task-bug-reproduction-success task-identify-harness-labels-done task-identify-issue-type-done
難易度 2/5 1〜3時間 初心者へのやさしさ 62/100
メンテナーはふだん 1 日以内に返信
-
[Bug]: Web chat input doesn't regain focus after a reply finishes対応中かも @GaijinSystems が今日担当しました。 オープン
難易度 2/5 1〜3時間 初心者へのやさしさ 76/100
zeroclaw-labs/zeroclaw#11658 ·
メンテナーはふだん 2 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 75/100
babylonlabs-io/babylon-toolkit#2711 ·
メンテナーはふだん 1 日以内に返信