Commit Graph
6 Commits
Author SHA1 Message Date
6cad158a17 security: reject PATH_TO_CLAUDE_CODE_EXECUTABLE with control characters (#1185)
dirname() preserves embedded newlines, so a value like
`/usr/bin/claude\n/attacker/path` writes two lines to GITHUB_PATH,
injecting an attacker-controlled directory into PATH for all subsequent
workflow steps.

Validate the input immediately after reading it and throw if it
contains any control characters (0x00-0x1f, 0x7f). This is fail-closed
rather than silent stripping — a path with control characters is always
misconfigured or malicious.

Fixes #1160

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-05 17:26:08 -07:00
0f1fe5ef85 fix: forward MCP_TIMEOUT, MCP_TOOL_TIMEOUT, MAX_MCP_OUTPUT_TOKENS to action step (#1162)
These three env vars are read directly from process.env by the Claude CLI
subprocess to configure MCP server behavior. Users setting them in their
workflow had no reliable way to make them reach the CLI:

- Job-level env: shadowed by the step's explicit env: block
- Step-level env: on the calling workflow step is not inherited by
  composite action steps
- GITHUB_ENV from a prior step: same shadowing problem
- settings input: writes to ~/.claude/settings.json, not process.env

The fix is to add explicit ${{ env.VAR }} passthrough lines for all three
vars, matching the existing pattern already used for OTEL_*, AWS_*, and
Vertex configuration (lines 271-317). No TypeScript changes are needed;
the forwarding chain in parse-sdk-options.ts is already correct.

Fixes #1152

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-05 10:37:03 -07:00
5150ea9643 fix: snapshot PR's .claude/ to .claude-pr/ before security restore (#1172)
When a PR modifies files under .claude/, the security restore in
restoreConfigFromBase() overwrites them with the base branch version —
correct for execution safety, but it means review agents never see what
the PR actually changes.

Before deleting the PR-controlled .claude/ tree, copy it to .claude-pr/.
Review agents can read .claude-pr/ to inspect the PR's hooks, MCP
configs, settings, and CLAUDE.md without those files ever being executed.
The snapshot is taken before the security delete so it captures the full
PR-authored version.

Fixes #1134.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-04 20:47:27 -07:00
f328a5c889 fix: prevent hang in restoreConfigFromBase on repos with .gitmodules (#1166)
When a PR head contains `.gitmodules`, git's default
`fetch.recurseSubmodules=on-demand` config causes `git fetch` to attempt
submodule object fetches. In CI (no credentials), this blocks indefinitely
waiting for auth — producing ~4-hour hangs reported in #1088.

Two changes, both defence-in-depth:

1. Delete SENSITIVE_PATHS *before* fetching. The attacker-controlled
   `.gitmodules` is absent during the network operation, so git never
   sees a submodule config to follow regardless of git settings.

2. Pass `--no-recurse-submodules` to the fetch. Suppresses submodule
   fetching explicitly, independent of any git config on the runner.

The original order (fetch-then-delete) was a brief window where
`.gitmodules` from the PR head could influence the fetch. Reordering
also tightens the security property: if `git checkout` below fails, the
attacker-controlled file is already gone rather than present during fetch.

Fixes #1088.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-04 20:21:28 -07:00
b15d4751a6 fix: allow # in branch names for PR checkout and base restore (#1167)
`validateBranchName` used a strict whitelist that excluded `#`,
causing the action to fail on PRs from branches like `put-back-arm64-#2`
with "Invalid branch name" — even though the branch already exists in
git and `#` is permitted by git-check-ref-format.

The validation was designed to prevent command injection. However, every
git call in the action uses `execFileSync`, which bypasses the shell
entirely and passes arguments directly to the kernel's execve. There is
no shell to interpret `#` as a metacharacter, so the strict whitelist was
over-blocking valid names with no security benefit.

Add `#` to the whitelist pattern, and update the JSDoc and error message
to reflect the allowed character set.

Fixes #1137.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-04 20:17:46 -07:00
d5db8208f9 fix: restore ripgrep execute bits after bun install --production (#1163)
bun install --production strips execute bits from vendored binaries
(bun bug). The Claude Agent SDK ships rg binaries in:

  node_modules/@anthropic-ai/claude-agent-sdk/vendor/ripgrep/
    {x64,arm64}-{linux,darwin}/rg
    {x64,arm64}-win32/rg.exe

After bun --production, all of these lose +x, causing EACCES when the
SDK tries to spawn ripgrep. The fix is a targeted find(1) that restores
+x on the rg binaries immediately after bun install.

Design notes:
- -type f excludes symlinks (symlink attack safety, no || true needed)
- -name "rg" naturally excludes rg.exe on Windows (find returns nothing,
  chmod never called — safe and correct on all platforms)
- .node audio-capture files use dlopen, not exec — no +x needed there
- Fails loudly if the binary path is missing (no || true) so a SDK
  packaging change is immediately visible rather than silently broken

Fixes #1140

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-04 20:15:31 -07:00