fix(core): evaluate permissions for commands the scanner cannot decompose - #49954

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

The following comment was made by an LLM, it may be inaccurate:

Note on the needs:issue label — I think it's a false positive on v2 PRs rather than a missing link.

The body does contain Closes #49948, and running pr-standards.yml's own fallback against this exact body gives:

section matched:    true
issueContent:       "Closes #49948"
hasBodyIssueRef:    true

so hasLinkedIssue should be true and removeLabel('needs:issue') should run. check-standards does report success, and the label still lands ~6s after the run starts. Same on #48968 and #48969, which are also v2-based — consistent with the closingIssuesReferences caveat already noted in the workflow, where the GraphQL count is 0 for non-default-branch PRs and only the body fallback can clear it. Happy to open a separate issue if that's useful.

Independent verification on current dev (2fa3363): confirmed at the scanner level. ShellTool.collect walks commands(root) = tree.rootNode.descendantsOfType("command"), and a redirect-only statement produces no command node, so scan.patterns stays empty and ask() returns before any permission assertion (the scan.patterns.size === 0 early-return in packages/opencode/src/tool/shell.ts). Same counts as the table above on both the bash and PowerShell parsers.

One linkage worth keeping in scope during review: for the compound case (echo foo; > out.txt) patterns are non-empty via the echo command, so the residual gap there is redirect-text attachment on decomposed commands — that's the #52083 / #52180 side of the family. This PR plus that work would close the class end to end.

I hit the same fail-open from the PowerShell side (#54393: a bare -- in a pwsh command parses to zero command nodes) and have a tool-level PowerShell regression test for it that works against this fix unchanged — dropping it here in case it's useful. It exercises both scanners end-to-end through the shell tool: the ask path (request must carry the raw command) and a deny-all rule (must fail with "Permission denied: shell"; pre-fix the command runs with no request at all):

test(`PowerShell ${portable ? "native" : "legacy"}: a bare -- command cannot skip the shell permission check`, () =>
  withScanner(
    portable,
    (registry, directory) =>
      Effect.gen(function* () {
        const command = "git rev-parse HEAD --"
        const marker = path.join(directory.active, "marker")
        const asked = yield* runPermissionCommand(registry, command, marker, ["reject"])
        expect(asked.requests).toMatchObject([
          {
            action: "shell",
            resources: [command],
            save: portable ? ["git rev-parse *"] : [command],
          },
        ])
        expect(Exit.isFailure(asked.exit)).toBe(true)
        if (Exit.isFailure(asked.exit))
          expect(
            asked.exit.cause.reasons.some(
              (reason) => Cause.isDieReason(reason) && reason.defect instanceof Permission.DeclinedError,
            ),
          ).toBe(true)

        const agents = yield* Agent.Service
        yield* agents.transform((editor) =>
          editor.update(toolIdentity.agent, (agent) => {
            agent.permissions = [{ action: "shell", resource: "*", effect: "deny" }]
          }),
        )
        const denied = yield* runPermissionCommand(registry, command, marker, [])
        expect(denied.requests).toEqual([])
        expect(denied.exit).toMatchObject({
          _tag: "Success",
          value: { status: "error", error: { message: expect.stringContaining("Permission denied: shell") } },
        })
        expect(yield* Effect.promise(() => Bun.file(marker).exists())).toBe(false)
      }),
    pwsh ?? "pwsh",
  ))

It slots into the PowerShell for (const portable of [false, true]) loop in packages/core/test/tool-shell.test.ts. Verified it fails on dev (no request fires) and passes with this PR's fallback; it needs pwsh on the runner for shell resolution, same guard the existing PowerShell tests there already use.

Automated PR Cleanup

Thank you for contributing to opencode.

Due to the high volume of PRs from users and AI agents, we periodically close older PRs using automated criteria so maintainers can focus review time on the most active and community-supported contributions.

This PR was closed because it matched the following cleanup criteria:

• The PR was created more than 1 month ago
• The PR had fewer than 2 positive reactions
• Positive reactions are counted as thumbs-up, heart, celebration, or rocket reactions on the PR

PRs created within the last month are not affected by this cleanup.

If you believe this PR was closed incorrectly, or if you are still actively working on it, please leave a comment explaining why it should be reopened. A maintainer can review and reopen it if appropriate.

Thanks again for taking the time to contribute.