fix(cli): do not report tip behind-count on diverged branches

[smfworks]

Summary

On a diverged feature branch, git rev-list --count HEAD..origin/main can still print a tiny tip count (e.g. 1 commit behind) even when HEAD and origin/main are mutual non-ancestors and the trees differ by hundreds/thousands of files. The banner and hermes --version then recommend a casual hermes update as if it were a small fast-forward.

Fix

• Add UPDATE_DIVERGED = -2 sentinel
• After computing the tip behind-count on full clones, require git merge-base --is-ancestor HEAD origin/main before reporting an exact behind number
• If not a fast-forward ancestor: return UPDATE_DIVERGED instead of the misleading tip count
• Banner copy: branch diverged from origin/main — not a fast-forward; review before hermes update (switches to main / may stash WIP)
• Version printer: same clear non-FF messaging (also surfaces UPDATE_AVAILABLE_NO_COUNT)

Shallow clones and pure fast-forward behind counts are unchanged.

Out of scope (follow-ups from #68484)

Desktop release/win-unpacked lock planner / pack-to-release-verify path — separate PR.

Test plan

• Full clone FF path still returns exact behind count
• Diverged branch with tip count 1 → UPDATE_DIVERGED (not 1)
• Existing update-check suite green (16 tests)
• Expired-cache call count updated for ancestry probe

Closes #68484

[teknium1]

Thanks for tracing the misleading full-clone count; current main still returns HEAD..origin/main directly at hermes_cli/banner.py:255-263, so the premise is valid.

Problems

• hermes_cli/banner.py:273 only checks ancestry when behind > 0. A diverged branch can have a zero HEAD..origin/main count; this patch would return 0, and the version path consequently prints Up to date (hermes_cli/main.py:4710-4718). Please check ancestry for every successful full-clone count and add the zero-count divergence test.
• UPDATE_DIVERGED changes a shared result contract. tui_gateway/server.py:4738-4742 forwards it, but Ink renders an update row only for update_behind > 0 (ui-tui/src/components/branding.tsx:467-479). Handle the new state there as well; the dashboard contract currently documents only -1 (hermes_cli/web_server.py:4144-4150).

Suggested changes

• Carry the divergence state through all consumers of check_for_updates() and cover each affected rendering contract.

Automated hermes-sweeper review.

[teknium1]

This leaves a diverged branch with HEAD..origin/main == 0 classified as up to date. The stated sentinel contract is ancestry-based, so run the ancestry check for every successful full-clone count and add the zero-count divergence regression case.

[smfworks]

Sweeper response — salvageability=medium

Head: (see latest push)

1. Ancestry check for every full-clone count (including behind==0 divergence).
2. UPDATE_DIVERGED = -2 surfaced in banner CLI, TUI branding, and dashboard web payload messaging.
3. Focused divergence regressions (positive behind + zero behind).

Verification

pytest tests/hermes_cli/test_update_check.py -q
4 passed · ruff clean

[smfworks]

Verification Summary (adversarial 6-check — PASS after sibling fix)

Head: (latest push)

1. Diff integrity: PASS
2. Blast radius: PASS — banner + CLI version + TUI branding + web_server + desktop version-status/overlay
3. Siblings: PASS — desktop was missing; fixed this pass (behind === -2)
4. Edges: PASS — behind>0 diverge and behind==0 diverge
5. Squash safety: PASS
6. Hostile reviewer: PASS after desktop fix — tip count no longer possible as +-2

Tests: Python 4 passed · ruff clean · desktop vitest not runnable in bare worktree (no node_modules); pure TS change + unit test added
Finding fixed this adversarial pass: desktop sibling for UPDATE_DIVERGED

[GottZ]

This was generated by AI during triage.

Summary

One PR, #68677, addresses #68484. Its diff replaces misleading tip-only counts with ancestry-based divergence detection across CLI, TUI, web, and Desktop consumers, including positive- and zero-count regressions, while the issue’s diff-shortstat reporting and Windows Desktop/venv lock preflight are not implemented.

Related pull requests

• fix(cli): do not report tip behind-count on diverged branches #68677 best fix — (+171/-11) — n/a; recorded best fix: detects both positive- and zero-count divergence and propagates UPDATE_DIVERGED through the banner, version output, TUI, web payload, and Desktop UI with focused tests, but does not add diff-shortstat or Windows lock preflight. The contributor’s COMMENTED keep_open review identified the zero-count and shared-consumer gaps; the current diff addresses those specific objections.

Suggested consolidation

Keep #68677 open with a salvage path: preserve its tested divergence detection and end-to-end consumer handling as the best existing partial fix, while requiring the closes relationship to account for the still-open diff-shortstat and Windows Desktop/venv lock-preflight scope, either in this PR or concrete follow-up work.

Complex graph

flowchart LR
    classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
    classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
    classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef best stroke-width:3px,stroke:#b45309
    classDef target stroke-width:3px,stroke:#4338ca
    I68484(["issue #68484 (open)"])
    P68677["PR #68677 (open)"]
    P68677 -->|best fix| I68484
    class I68484 open
    class P68677 open
    class P68677 best
    class P68677 target
    click I68484 "https://github.com/NousResearch/hermes-agent/issues/68484"
    click P68677 "https://github.com/NousResearch/hermes-agent/pull/68677"

Loading

Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed (state tag in the node label).

Cross-PR triage: Reviewed 1 pull request and 1 issue in this complex. Each diff was read against this issue; Assessment working set: 14 kB of PR diffs, 7 kB of issue/PR text, 3 kB of discussion (5 comments), 2 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.

[smfworks]

Stewardship ping — verified this fix is still needed on current main:

• No ancestry-based divergence detection found on main
• check_for_updates() still uses rev-list --count HEAD..origin/main which reports misleading tip counts on diverged branches
• Sweeper previously marked salvageability=medium

The PR replaces tip-only counts with ancestry-based divergence detection across CLI, TUI, web, and desktop consumers (including UPDATE_DIVERGED = -2 sentinel). Adversarial 6-check passed. Ready for review.

[smfworks]

Stewardship follow-up — rebased onto current main (6564f319)

Cherry-pick clean, 8 files changed, 171 insertions, 11 deletions.

The divergence signal is still absent from main. No UPDATE_DIVERGED constant or diverged-branch detection exists in cli.py. When the local branch diverges from upstream (not behind, not ahead — diverged), the tip still reports a misleading behind-count. This PR surfaces -2 as UPDATE_DIVERGED across banner CLI, TUI branding, and dashboard web payload.

Test status: tests/hermes_cli/test_update_check.py — 4 passed on rebased branch.

Ready for review.

[smfworks]

Restored onto current main as a focused 3-file desktop/TUI divergence-signal diff. Previous tip had accumulated unrelated fleet files.

[teknium1]

Partial overlap heads-up: PR #86257 (merged) landed the behind-count class fix — the SSH-official and shallow paths no longer fabricate counts, and version-status.ts gained an updateAvailable count-free rendering path for both targets. That covers the "don't render a wrong tip count" half of this PR. The diverged-specific signal (behind === -2 → "(diverged)" label + tooltip) is NOT covered — divergence still isn't distinguished from "behind". If you rebase onto current main and rescope to the diverged sentinel plumbing, that part remains reviewable on its own.