release: 2.3.0 - #25
Merged
Merged
Conversation
A rule that opens with `*` has nothing before its wildcard, and what
precedes the wildcard is what limits a rule. So it approves a shape, and
any program can wear it. rodekruis/qualitative-feedback-analysis, a
public Netherlands Red Cross repository, carries `Bash(* --version)` and
`Bash(* --help *)`, and both read as courtesies for version banners.
Measured on Claude Code 2.1.238 with a file-creating probe, chosen so the
read-only classifier could not approve it on its own: under
`Bash(* --version)`, `bash -c "touch <marker>" --version` ran and the
marker appeared. With no rules present it did not run, so the grant is
the rule and not the environment. Without the flag it did not run, and
neither did the same command with a word after it.
`Bash(*)` is the honest spelling of the same reach and is reported
plainly. Deny rules keep their leading stars without comment: in a
344-file sample most leading-star rules were denies such as
`Bash(*xmrig*)`, and breadth in a deny runs in the safe direction. Across
12,556 rules the check fires 33 times.
Also fixes a false sentence the word-boundary check printed on those same
deny rules ("every program whose name starts with `*xmrig`"), and returns
a doc comment orphaned in 2.2.0 to its function.
The version bump rewrote every @agentfile/* range in the workspace, including the CLI dependency on @agentfile/ui. That package is published separately at 0.1.0-beta.0, so ^2.3.0 resolves to nothing and every CI job failed at npm install with ETARGET.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A rule that opens with
*has nothing before its wildcard, and what precedes the wildcard is what limits a rule. So it approves a shape, and any program can wear it.The case that found it: rodekruis/qualitative-feedback-analysis, a public Netherlands Red Cross repository, carries
Bash(* --version)andBash(* --help *). Both read as courtesies for version banners. agentfile 2.2.0 said nothing about either.Measured, not inferred
Claude Code 2.1.238, with a file-creating command as the probe so the read-only classifier could not approve it on its own, and the marker file as proof of execution:
bash -c 'touch M' --versionunderBash(* --version)... --version extrabash -c 'touch M' --help extraunderBash(* --help *)bash -c 'touch M'underBash(*)The tail still has to match, which is exactly what makes the shape read narrower than it is. An earlier pass of this measurement used
echoand produced a contradictory row: Claude Code's read-only classifier can approve such a command by itself, so the probe was rebuilt around a write.What it reports
Bash(* --version)is reported as an arbitrary-command grant.Bash(*)is the honest spelling of the same reach and is reported plainly, since the reader's remaining guards are deny rules, hooks and the sandbox. Where the tail names a program, as inBash(*vitest*), the suggestion is the rule the author meant:Bash(vitest *).Scope, from the corpus
Deny rules keep their leading stars without comment. In a 344-file sample most leading-star rules were denies such as
Bash(*xmrig*)andBash(* | sh), and breadth in a deny runs in the safe direction. Across 12,556 rules the new check fires 33 times: 4 tail-constrained rules in 2 repositories, and 29 spellings ofBash(*).Also in this release
The word-boundary check printed a false sentence on those same deny rules:
Bash(*xmrig*)drew "also matches every program whose name starts with*xmrig", and no program is named*xmrig. It now stands down where a rule opens with a wildcard, which removes 29 such sentences across the sample. A doc comment orphaned from its function in 2.2.0 sits with it again.1042 tests pass, lint and typecheck are clean, and the built CLI on the rodekruis clone reports 4 warnings: the two
git * main|masterrules and the two new ones.