audit(unresolved): batched fork-PR audit reverts — 6 PRs (#1189, #1192, #1188, #1194, #1193, #1190) by Kyzcreig · Pull Request #1230 · ANG-Ventures/hermes-agent

Kyzcreig · 2026-09-26T06:21:47Z

Fork-PR audit batch for the unresolved area (card t_7c5e08a5, campaign t_f201acde, lead t_03e35f0e; rulings in #1128 / #1186).

It replaces 6 separate queue entries with 1. Each separate squash invalidated every group behind it, and the members' overlapping files caused the 2026-09-25 UNMERGEABLE wave. From now on each audit area gets at most 1 open queue entry.

PR	Title	Card
#1189	revert(tui_gateway): drop the heavy session-read admission gate (#215, #441)	t_7c17872a
#1192	revert: drop #358 empty prompt.submit guard + no-op model-switch short-circuit	t_28b13602
#1188	Revert #620: strip relay routing prefix before model-listing lookup (audit DROP)	t_111298c6
#1194	revert: drop #894 residual pasteable-hint sites (fork-PR audit DROP)	t_ceb111f9
#1193	revert(recap): drop backtick-wrap guard on dead session_recap module (audit DROP nopr:8a8b81638c)	t_08d470c3
#1190	revert(skills): drop write-time skill-hygiene guard A3 (`8dcc696` + `08fc3af`) — audit DROP	t_cad124fd

How it was built: branched from origin/main (cebd579), then git merge --no-ff of each member PR head in order. No revert content was rewritten. Each member's own conflict resolutions come through unchanged.

Identity check: every file a member touches is byte-identical to that member's head, with two exceptions. (a) Files two members share, or files main changed after a member's base: the +/- lines were compared as a multiset against the union of the members' own diffs, with 0 missing and 0 extra. (b) The file deletions noted below. No file outside the members' changes differs from main. Total: 24 files, +33/-2093.

#1194 carried "Do not merge without Apollo pass" — that condition carries over to this batch. Merge conflicts vs main (#1194 was 62 commits behind): staging/ was deleted on main by #1158, so it stays deleted; tests/test_residual_cli_hint_pasteable.py is deleted as #1194 intends (main's only edit to it dropped the staging group).

Local verification (narrow, repo venv py3.11): py_compile passes on every changed .py file. On the unresolved batch (the server.py overlap between #1189 and #1192), pytest tests/test_tui_gateway_server.py tests/test_tui_gateway_ws.py tests/hermes_cli/test_model_validation.py gives 739 passed. The full suite and matrix are left to CI.

The member PRs will be closed and pointed here once this PR is open.

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

…as a script → CI collection abort)" This reverts commit 08fc3af.

This reverts commit 8dcc696.

…lookup (#620)" This reverts commit ac92718. Fork-PR audit verdict DROP (t_63023f77, docs/plans/fork-pr-audit/UNRESOLVED.md): 0/34,120 sessions across 12 profile state.db (2026-05-08..2026-09-26) use a name:<sub>/<model> model string; 0 config.yaml/cron jobs use one; 0 'could not verify name:' log lines. Upstream validate_requested_model has no equivalent strip. Warning-text fix for a model-string form nobody uses. Conflict in tests/hermes_cli/test_model_validation.py resolved by deleting only TestSubRoutingPrefixStripped; later-added classes kept. Verified: test-gate pytest tests/hermes_cli/test_model_validation.py -> 64 passed. Card: t_111298c6

…#441) Reverts 66d98d3 (#215) and its re-land 0986938 (#441) as one unit, per the fork-PR audit UNRESOLVED ruling (DROP, t_63023f77 / t_7c17872a). Removed: - hermes_cli/session_db_heavy_gate.py (SessionDBHeavyReadBusy, the slot, stats, lazy dashboard.heavy_read_max_concurrency read) - tui_gateway/ws.py: gate import + _dispatch_ws_request; WS requests go back to asyncio.to_thread(server.dispatch, ...) - tui_gateway/server.py: _SESSION_DB_HEAVY_METHODS, is_session_db_heavy_method, backend_busy_error, handle_request_bound (_err's data= param kept: methods_prompt/methods_bot_relay use it) - hermes_cli/web_server.py: gate import, 503 shed handler and the /api/status session_db_heavy_reads field; _session_db_read(heavy=True) returns to the pre-#215 per-loop asyncio.Semaphore(2) - config_defaults: dashboard.heavy_read_max_concurrency - gate tests in tests/test_web_server_sessiondb_eventloop.py and tests/scripts/test_preyield_permit_release.py; the guard script's live-site list. The guard sweep keeps a live site (gateway/turn_admission.py) and its named coverage test. Why: 0 'session_db_heavy_read' log lines across ~/.hermes/logs and profiles/*/logs (no read ever queued >= 1 ms); 473/474 loc across 5 files conflicting in all 3 upstream syncs; upstream bounds reads in its SessionDB read pool instead. Verified: test-gate (ACE-AI) pytest tests/test_web_server_sessiondb_eventloop.py tests/scripts/test_preyield_permit_release.py tests/test_tui_gateway_ws.py tests/tui_gateway/test_ws_keepalive.py -> 85 passed. ruff --select F on the touched files: no new findings vs origin/main.

Reverts e1de835. Fork-PR audit UNRESOLVED ruling (t_63023f77, #1186): the trigger is whitespace in a home/interpreter/skill-category/plugin path; measured 0 occurrences on both fleet hosts, so the hints never needed quoting here. The gateway/run.py hunk conflicted in all 3 upstream syncs. Kept: #889's hermes_cli.cli_hint.hint_value and its callers (KEEP row). gateway/run.py conflict resolved by keeping current main's _skill_slug_index loop and dropping only the hint_value() wrap. Card: t_ceb111f9

…t-circuit Reverts 2850311 (#358) per fork-PR audit DROP ruling (t_63023f77, docs/plans/fork-pr-audit/UNRESOLVED.md). The prompt.submit guard now lives in tui_gateway/methods_prompt.py after the parity sync split, so it is removed there; the _apply_model_switch no-op block is removed from server.py; the 4 tests the commit added are removed from tests/test_tui_gateway_server.py. Why: the path carries ~no traffic (8 source=tui sessions of 15,265, 0 empty user rows from tui) and the looping client (fork desktop reconnect) is retired by D9. 237 loc across files that conflict every sync. Reopen if a desktop empty-submit loop recurs.

… auto-attach" This reverts commit 8a8b816. Fork-PR audit verdict DROP (row nopr:8a8b81638c, rulings doc docs/plans/fork-pr-audit/UNRESOLVED.md, PR #1186): build_recap, the only public entry of hermes_cli/session_recap.py, has 0 non-test callers on fork main (git grep over *.py/*.ts/*.tsx/*.yaml at 702addb) and on upstream. The guarded path cannot run, so the guard fires 0 by construction. Test-file conflict (parity merge aa27fd8 reshaped the file) resolved by taking main's version and deleting only the two tests 8a8b816 added. Card: t_08d470c3

…215-441)

…358)

…620)

Conflicts vs main (PR was 62 behind): staging/ deleted on main by #1158 -> stays deleted; tests/test_residual_cli_hint_pasteable.py deleted per #1194 (main's edit only dropped the staging group).

…aths to sto)

…8dcc69611c)

blacksmith-sh · 2026-09-26T06:24:44Z

Found 1 test failure on Blacksmith runners:

Failure

Test	View Logs
`TestPinTransition/test_cache_busting_signature_reflects_pin_peer_name`	View Logs

^{Need help on this PR? Tag @codesmith-bot with what you need.}

ang-fleet-lander · 2026-09-26T15:39:37Z

🤖 merged-by: apollo · lane: discord · gate: BYPASS: FR paused by Ace 2026-09-22; Argus off card review (Ace 09-24 13:08); gate = handoff + CI green + Apollo read (review-of-record on card) · why: batched audit reverts unresolved

ang-prism · 2026-09-27T11:48:25Z

FleetReview

Post-merge review (fleetreview:post-merge override): this reviewed the merge commit against its first parent — the bytes that already shipped. It is not a pre-merge gate pass.

Confidence: 2/5

Findings

P1 tools/skill_manager_tool.py:1213 — Unprotected placement
P1 agent/skill_utils.py:118 — Queue notes
P1 hermes_cli/models.py:6984 — Incorrect lookup
P2 hermes_cli/web_server.py:3269 — Unbounded waiting
P2 tools/self_repo_guard.py:736 — Broken clone command
P2 hermes_cli/web_server.py:3254 — Stale semaphore
P1 hermes_cli/session_recap.py:302 — File disclosure
P1 tui_gateway/server.py:6844 — No-op switch
P1 tui_gateway/methods_prompt.py:335 — Blank submissions
P2 tui_gateway/ws.py:490 — Ungated scans
P2 plugins/platforms/whatsapp/adapter.py:612 — Unquoted paths
P1 hermes_cli/plugins_cmd.py:1103 — Shell-quote plugin sources in copy-paste install commands
P1 tests/hermes_cli/test_session_recap.py:49 — Removed Security Test
P1 tests/test_residual_cli_hint_pasteable.py:384 — Pinned-plugin update hint prints an executable, unquoted source

FleetReview provenance · models: B=gpt-6-sol, C=claude-code-opus-5-5, D=grok-4.6, F=gpt-6-sol · cost: $4.48 · duration: 13m 10s · rounds: 1 · files examined: 24

…s (t_9c9e757c) FleetReview retro-backfill 2026-09-27, class C4. Each finding re-verified at 5a9d284; one RED-on-base / GREEN test per confirmed instance. - #813 agent/quota_registry_gate.py: missing/unparseable observed_at now fails open (docstring contract), no longer skips a working fallback. - #1017 cron/lifecycle_guard.py: mask allowlist refuses module callables passed as values (json.dumps(p, default=subprocess.run), sorted(key=...)). - #1019 agent/lsp/manager.py: slot-dir OSError in _has_capacity -> run without LSP instead of raising out of enabled_for()/write_file. - #1026 cron/scheduler.py: owner-deadman exemption scoped to the host it owns; a note naming only another DOWN host is gated. - #1229 scripts/ci/live_comment.py: comment lookup/update network errors no longer kill the poller; a failed post is retried next poll. - #1230 plugins_cmd.py: pinned-plugin remedy shell-quotes the recorded source (CLI + dashboard). - #1238 tools/mixture_of_agents_tool.py: reference responses move from the aggregator system message to a tagged data block in the user message. Verified: test-gate narrow pytest; new tests 13 RED on base, all GREEN with fix; neighbouring suites (lifecycle_guard x4, host_down_gate, quota wiring, plugins_cmd, fallback_policy, fork toolsets) pass.

…s (t_9c9e757c) FleetReview retro-backfill 2026-09-27, class C4. Each finding re-verified at 5a9d284; one RED-on-base / GREEN test per confirmed instance. - #813 agent/quota_registry_gate.py: missing/unparseable observed_at now fails open (docstring contract), no longer skips a working fallback. - #1017 cron/lifecycle_guard.py: superseded by #1348 (6662bbd) on main; this slice's 5 callback cases pass against it, so the edit was dropped on rebase. - #1019 agent/lsp/manager.py: slot-dir OSError in _has_capacity -> run without LSP instead of raising out of enabled_for()/write_file. - #1026 cron/scheduler.py: owner-deadman exemption scoped to the host it owns; a note naming only another DOWN host is gated. - #1229 scripts/ci/live_comment.py: comment lookup/update network errors no longer kill the poller; a failed post is retried next poll. - #1230 plugins_cmd.py: pinned-plugin remedy shell-quotes the recorded source (CLI + dashboard). - #1238 tools/mixture_of_agents_tool.py: reference responses move from the aggregator system message to a tagged data block in the user message. Verified: test-gate narrow pytest; new tests 13 RED on base, all GREEN with fix; neighbouring suites (lifecycle_guard x4, host_down_gate, quota wiring, plugins_cmd, fallback_policy, fork toolsets) pass.

…s (t_9c9e757c) (#1353) FleetReview retro-backfill 2026-09-27, class C4. Each finding re-verified at 5a9d284; one RED-on-base / GREEN test per confirmed instance. - #813 agent/quota_registry_gate.py: missing/unparseable observed_at now fails open (docstring contract), no longer skips a working fallback. - #1017 cron/lifecycle_guard.py: superseded by #1348 (6662bbd) on main; this slice's 5 callback cases pass against it, so the edit was dropped on rebase. - #1019 agent/lsp/manager.py: slot-dir OSError in _has_capacity -> run without LSP instead of raising out of enabled_for()/write_file. - #1026 cron/scheduler.py: owner-deadman exemption scoped to the host it owns; a note naming only another DOWN host is gated. - #1229 scripts/ci/live_comment.py: comment lookup/update network errors no longer kill the poller; a failed post is retried next poll. - #1230 plugins_cmd.py: pinned-plugin remedy shell-quotes the recorded source (CLI + dashboard). - #1238 tools/mixture_of_agents_tool.py: reference responses move from the aggregator system message to a tagged data block in the user message. Verified: test-gate narrow pytest; new tests 13 RED on base, all GREEN with fix; neighbouring suites (lifecycle_guard x4, host_down_gate, quota wiring, plugins_cmd, fallback_policy, fork toolsets) pass. Co-authored-by: ang-fleet-workers[bot] <333956806+ang-fleet-workers[bot]@users.noreply.github.com>

…m undo, shlex remedy, root-equality noqa - plugin_dev._load_model_provider: undo registry additions via provider_seam._restore (fork additive facades forbid pop/clear). - plugins_cmd_update: re-thread fork _shell_quoted_source (#1230 C4) that upstream's extraction of cmd_update/dashboard_update dropped. - root-equality lint: annotate 7 upstream-new relative_to sites (all FILE paths or non-derived rmtree targets) with reasoned noqa. - test_ollama_cloud_auth: seed _DIRECT_ALIAS_LOADED so the hot-reload test exercises the loader-owned-cache path (upstream NousResearch#16767 keeps caller-seeded dicts). - test_oneshot_reasoning_and_tier: FakeAgent.run_conversation accepts conversation_history. Verified (e45-pt-M2.sh / e45-pt-M2-314.sh via test-gate): test_ollama_cloud_auth + test_direct_alias_reload + test_model_alias_credentials: 79 passed test_oneshot_reasoning_and_tier: 9 passed test_plugin_dev + test_provider_discovery_in_progress: 10 passed (also 1 passed on py3.14) test_root_equality_containment_lint: 3 passed test_plugin_install_ref (py3.14, uv on PATH): 33 passed

Kyzcreig and others added 17 commits September 25, 2026 19:21

Revert "fix(test): rewrite A3 guard test as a proper pytest module (w…

4079997

…as a script → CI collection abort)" This reverts commit 08fc3af.

Revert "feat(skills): write-time skill-hygiene guard (A3)"

9c47337

This reverts commit 8dcc696.

Merge branch 'main' into audit/unresolved/revert-358

6eefcc1

Merge branch 'main' into audit/unresolved/revert-8dcc69611c

e56ec9b

Merge branch 'main' into audit/unresolved/revert-215-441

593e8db

Merge branch 'main' into audit/unresolved/revert-620

8cd8b9d

batch: merge #1189 (Merge branch 'main' into audit/unresolved/revert-…

7461616

…215-441)

batch: merge #1192 (Merge branch 'main' into audit/unresolved/revert-…

04c1292

…358)

batch: merge #1188 (Merge branch 'main' into audit/unresolved/revert-…

7605d8d

…620)

batch: merge #1194 (revert #894 residual pasteable-hint sites)

fb36a42

Conflicts vs main (PR was 62 behind): staging/ deleted on main by #1158 -> stays deleted; tests/test_residual_cli_hint_pasteable.py deleted per #1194 (main's edit only dropped the staging group).

batch: merge #1193 (Revert "fix(recap): backtick-wrap Files touched p…

80f36ea

…aths to sto)

batch: merge #1190 (Merge branch 'main' into audit/unresolved/revert-…

ce087ba

…8dcc69611c)

Merge remote-tracking branch 'origin/main' into audit/batch/unresolved

c0552cc

ang-fleet-lander Bot added this pull request to the merge queue Sep 26, 2026

Merged via the queue into main with commit 4e9dbb7 Sep 26, 2026
55 checks passed

ang-fleet-lander Bot deleted the audit/batch/unresolved branch September 26, 2026 16:16

This was referenced Sep 27, 2026

fix(guards): C4 backfill slice -- 6 confirmed guard/ownership findings (t_9c9e757c) #1353

Merged

fix(c6): fail closed on 28 fail-open / false-success gates (FleetReview backfill, t_6500a97a) #1362

Merged

This was referenced Sep 28, 2026

fix(kanban): C7 slice A durable-state fixes, 10 confirmed findings (t_60634825) #1373

Merged

fix: 7 FleetReview P2 backfill findings (t_f4ef102c) #1410

Merged

fix: 29 P2 backfill remainder-2 rows (B/C/E/F/G/I + A test proxies) (t_e9ca7d13) #1416

Merged

Navigation Menu

Repository navigation

audit(unresolved): batched fork-PR audit reverts — 6 PRs (#1189, #1192, #1188, #1194, #1193, #1190) - #1230

audit(unresolved): batched fork-PR audit reverts — 6 PRs (#1189, #1192, #1188, #1194, #1193, #1190)#1230
ang-fleet-lander[bot] merged 18 commits into
mainfrom
audit/batch/unresolved

Kyzcreig commented Sep 26, 2026 •

edited by blacksmith-sh Bot

Loading

Uh oh!

blacksmith-sh Bot commented Sep 26, 2026 •

edited

Loading

Uh oh!

ang-fleet-lander Bot commented Sep 26, 2026

Uh oh!

Uh oh!

ang-prism Bot commented Sep 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

Kyzcreig commented Sep 26, 2026 • edited by blacksmith-sh Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

blacksmith-sh Bot commented Sep 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Failure

Uh oh!

ang-fleet-lander Bot commented Sep 26, 2026

Uh oh!

Uh oh!

ang-prism Bot commented Sep 27, 2026

FleetReview

Findings

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Kyzcreig commented Sep 26, 2026 •

edited by blacksmith-sh Bot

Loading

blacksmith-sh Bot commented Sep 26, 2026 •

edited

Loading