# NousResearch/hermes-agent · Block prompt-injection data exfiltration via images/shells

## Open now

- 1 issue, 2 PRs open on GitHub.
- #129975: \[Feature\]: Only fetch reply image URLs that came from a tool, so a prompt injection cannot send data out through an image link
- PR #130023: fix(gateway): only fetch reply image URLs with tool provenance
- PR #134263: fix(security): kernel-enforce secret isolation for agent shells and close exfiltration paths

These items address security controls to prevent prompt-injection attacks from leaking data through image URL fetching or agent shell access. Changes include restricting image URL retrieval to tool-provenance sources and enforcing secret isolation in agent shells.

## TL;DR

- 3 member(s): 1 open issue(s), 0 closed issue(s), 2 open PR(s), 0 merged PR(s), 0 PR(s) closed without merging.
- No fix has been merged to main; one or more PRs were attempted (see below).
- **Recommended:** review the prior/competing PR(s) below. A fix was attempted but not merged; your case may extend or revive that work.

## Maintainer resolution verdict

- Internal-tracker verdict over this cluster's issue members: `unresolved`. This is distinct from the fix status above. Weigh it yourself.

## How this cluster was grouped

- Link evidence (`STRUCTURAL_SPANNING`): every member is connected by structural evidence: shared changed files, or a closing/superseded reference.
- Pairwise link density `0.67`: the share of member pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36. `1.00` means every pair is linked.

## Open related issues

- #129975: \[Feature\]: Only fetch reply image URLs that came from a tool, so a prompt injection cannot send data out through an image link

## Prior & competing PRs

- **#130023: fix(gateway): only fetch reply image URLs with tool provenance**. open
  - maintainer direction: category `silent_close`; implication `active`
  - files: `gateway/media_repair.py`, `gateway/platforms/base.py`, `gateway/run_turn.py`, `tests/gateway/test_background_command.py`, `tests/gateway/test_media_extraction.py`, `tests/gateway/test_media_resend_dedup.py`, `tests/gateway/test_platform_base.py`
- **#134263: fix(security): kernel-enforce secret isolation for agent shells and close exfiltration paths**. open
  - maintainer direction: category `silent_close`; implication `active`
  - files: `SECURITY.md`, `agent/file_safety.py`, `cli-config.yaml.example`, `cron/scheduler_script.py`, `docker/stage2-hook.sh`, `gateway/platforms/base.py`, `gateway/platforms/base_reply_images.py`, `hermes_cli/config_defaults.py`, `tests/cron/test_cron_multiplex_desktop_ticker_scope.py`, `tests/cron/test_cron_no_agent.py`, `tests/gateway/test_platform_base.py`, `tests/gateway/test_reply_image_exfil_gate.py`, `tests/tools/environments/test_gateway_code_dir_isolation.py`, `tests/tools/environments/test_landlock_plan_rules.py`, `tests/tools/environments/test_terminal_secret_isolation.py`, `tests/tools/test_browser_scheme_floor.py`, `tests/tools/test_code_kernel.py`, `tests/tools/test_hardline_normalization_bypass.py`, `tests/tools/test_local_env_blocklist.py`, `tests/tools/test_process_registry.py` (+19 more)

## Fix status

- No merged fix on record.
- Subsystem: `environments`