fix(contracts): abiba default model row + disk-gc access-method note #86

Merged

abiba-bot merged 2 commits from fix-litellm-health-keylist-20260913 into master 2026-09-13 19:09:21 +00:00

Conversation 0 Commits 2 Files Changed 2

+2 -1

abiba-bot commented 2026-09-13 18:58:07 +00:00

Owner

Copy Link

Copy Source

Two small contract corrections, both already verified by firstmate.

1. abiba-zulip-restore.prose.md — default model row

The table published Default model | deepseek-v4-pro while the row above it already stated provider syslog-harness. Live truth (/root/.pi/agent/settings.json): defaultProvider: syslog-harness, defaultModel: syslog-auto. deepseek-v4-pro is not a LiteLLM name at all — the harness catalogue is exactly syslog-auto, gpu-dense, gpu-vision, strix-moe — so a config following this row would authenticate (the key allowlist still carries it) and then fail as an unknown model. This is the last such claim in the repo; the source is now stated as settings.json.

Note: the primary pi session reaches DeepSeek through its own provider (auth.json), not through LiteLLM, so no gateway change is involved.

2. disk-gc-threat-response.prose.md — access-method note

Adds one line to the CT 100 probe-gap section: for kagentz (CT 105) use ssh root@kagentz rather than pct exec 105 (which reports loop0 at 59G while ssh sees the real 99G filesystem), and for the QEMU VM docker-vm (CT 109) use ssh root@192.168.68.7.

Disclosure: this second file is the same content that was in PR #85, which the pre-delivery review FAILED on the correct ground that documentation alone does not fix the reported defect. It is included here as the verified factual content only — the substantive scanner fix (deterministic reachability verdict, per-guest access method, per-field audit) remains tracked separately as disk-gc-probe-false-unreachable-20260913. Subsequent runs did report every host with real per-host values, which is why the content is worth landing, but reviewers should judge it as documentation, not as the fix.

Reviewers: the three-dot diff against master should be exactly these two files, one line each.

Two small contract corrections, both already verified by firstmate. ## 1. `abiba-zulip-restore.prose.md` — default model row The table published `Default model | deepseek-v4-pro` while the row above it already stated provider `syslog-harness`. Live truth (`/root/.pi/agent/settings.json`): `defaultProvider: syslog-harness`, `defaultModel: syslog-auto`. `deepseek-v4-pro` is not a LiteLLM name at all — the harness catalogue is exactly `syslog-auto`, `gpu-dense`, `gpu-vision`, `strix-moe` — so a config following this row would authenticate (the key allowlist still carries it) and then fail as an unknown model. This is the last such claim in the repo; the source is now stated as settings.json. Note: the primary pi session reaches DeepSeek through its own provider (auth.json), not through LiteLLM, so no gateway change is involved. ## 2. `disk-gc-threat-response.prose.md` — access-method note Adds one line to the CT 100 probe-gap section: for `kagentz` (CT 105) use `ssh root@kagentz` rather than `pct exec 105` (which reports loop0 at 59G while ssh sees the real 99G filesystem), and for the QEMU VM `docker-vm` (CT 109) use `ssh root@192.168.68.7`. Disclosure: this second file is the same content that was in PR #85, which the pre-delivery review FAILED on the correct ground that documentation alone does not fix the reported defect. It is included here as the verified factual content only — the substantive scanner fix (deterministic reachability verdict, per-guest access method, per-field audit) remains tracked separately as `disk-gc-probe-false-unreachable-20260913`. Subsequent runs did report every host with real per-host values, which is why the content is worth landing, but reviewers should judge it as documentation, not as the fix. Reviewers: the three-dot diff against master should be exactly these two files, one line each.

abiba-bot added 2 commits 2026-09-13 18:58:07 +00:00

fix: disk-gc-threat-response - correct access methods for kagentz and docker-vm ... c81cf5b6f0

- kagentz (105): use ssh root@kagentz (hostname), NOT pct exec 105 (shows loop0 59G, not real 99G)
- docker-vm (109): use ssh root@192.168.68.7 (correct QEMU VM access)
- abiba (100): use ssh root@abiba (hostname)
- syslog-api (116): use pct-run 116 (correct)

Verified:
  kagentz: 99G 8.9G 86G 10% /
  docker-vm: 158G 17G 135G 11% /
  abiba: 59G 13G 44G 23% /
  syslog-api: 40G 13G 25G 34% /

fix: change abiba default model from deepseek-v4-pro to syslog-auto ... 7906b2d52d

The abiba-zulip-restore.prose.md file incorrectly claimed the default model
was 'deepseek-v4-pro', but /root/.pi/agent/settings.json declares
'defaultModel: syslog-auto' and 'defaultProvider: syslog-harness'.

deepseek-v4-pro is not a servable LiteLLM model name for this fleet.
The only live models are: syslog-auto, gpu-dense, gpu-vision, strix-moe.

Note: The firstmate/pi session talks to DeepSeek through its own
provider (auth.json), NOT through LiteLLM, so no LiteLLM change is
needed to keep firstmate's deepseek usage working.

abiba-bot merged commit ba76f2c7d3 into master 2026-09-13 19:09:21 +00:00

Sign in to join this conversation.

Reviewers

No Reviewers

Labels

No items

No labels

Milestone

No items

No Milestone

Projects

Clear projects

No projects

Assignees

Clear assignees

abiba-bot jerome (Jerome Kwame Tabiri) kagentz-bot kwame (Kwame Tabiri) mumuni-bot (Mumuni 🦅 (Syslog Falcon)) tanko-bot theo

No Assignees

1 Participants

Notifications

Subscribe

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: SyslogSolution/prose-contracts#86