fix(insights): show cache tokens in /insights token breakdown

[liuhao1024]

What does this PR do?

The /insights report's total_tokens includes cache_read_tokens + cache_write_tokens, but the (in: / out: ) breakdown only showed input and output tokens. For Anthropic users with prompt caching enabled, cache tokens can dominate the total (e.g., 5.2M total but only 2.6K input / 88K output), making the in/out numbers appear swapped or wrong.

Root Cause

In agent/insights.py, the _compute_overview() method correctly sums all four token categories into total_tokens:

total_tokens = total_input + total_output + total_cache_read + total_cache_write

But both display methods (format_terminal and format_gateway) only showed input and output:

Tokens: 5,239,805 (in: 2,642 / out: 88,463)   # 5.15M cache tokens hidden!

Related Issue

N/A

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• See commit messages for detailed changes

How to Test

1. Run pytest tests/ -q — all tests should pass
2. Verify the specific scenario described above is resolved

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix(scope):, feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes (required for bug fixes, strongly encouraged for features)
• I've tested on my platform: macOS 26.4.1

Documentation & Housekeeping

• I've updated relevant documentation (README, docs/, docstrings) — or N/A
• I've updated cli-config.yaml.example if I added/changed config keys — or N/A
• I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture and workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A

[teknium1]

Thanks for identifying that the visible input/output figures do not sum to total_tokens.

Problems

• agent/insights.py:765 reintroduces cache-token display that merged PR #11477 deliberately removed as unreliable (commit e33cb65a9). Current main preserves that policy in tests/agent/test_insights.py:465-468, and format_gateway() still omits cache at agent/insights.py:871.

Suggested changes

• Please reconcile the cache-total proposal with #11477's reliability rationale before restoring the metric. If maintainers want a combined cache total despite that prior decision, make it an explicit presentation-policy change and update the current display-contract tests accordingly.

Automated hermes-sweeper review.

[teknium1]

This restores cache-token output that current main deliberately removed in #11477 (e33cb65a9) because cache metrics were unreliable. Current tests retain the no-cache display contract; please reconcile this change with that policy before reintroducing the value.

[GottZ]

This was generated by AI during triage.

Summary

Two PRs address the same token-breakdown mismatch: #4428 added conditional cache-token output to terminal and gateway reports, while #18632 independently restores that presentation and adds focused display-contract tests.

Related pull requests

• #4428 [merged] duplicate — (+8/-1) — historical implementation: #4428 conditionally exposed cache reads/writes so the visible breakdown matched total_tokens; it remains relevant as the merged reference implementation, although contributor review on #18632 states that commit e33cb65 later removed this display as unreliable.
• #18632 related — (+77/-1) — overlapping fix with tests: #18632 restores conditional cache totals in both formatters and adds five cache-display tests. The contributor review marked it keep_open but salvageability=none because the diff reverses the reliability policy established by #11477; it requires an explicit presentation-policy decision and corresponding display-contract tests before proceeding.

Duplicates

#4428 and #18632 implement essentially the same conditional cache-token presentation; #18632 additionally supplies targeted tests, while #4428 is the merged historical implementation later superseded by the removal cited in review.

Suggested consolidation

Keep #18632 open with a salvage path: retain its focused test coverage and conditional-formatting work only if the author first reconciles the proposal with #11477 and the reliability rationale behind commit e33cb65, then updates the tests to encode the resulting explicit presentation policy. Do not treat merged #4428 as the current fix: it is the duplicate historical implementation and was subsequently superseded by that removal.

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
    subgraph Dup4428 ["PRs duplicating each other"]
        P4428["PR #4428 (merged)"]
        P18632["PR #18632 (open)"]
    end
    class P4428 merged
    class P18632 open
    class P18632 target
    click P4428 "https://github.com/NousResearch/hermes-agent/pull/4428"
    click P18632 "https://github.com/NousResearch/hermes-agent/pull/18632"

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 2 pull requests and 0 issues in this complex. Each diff was read against this issue; Assessment working set: 7 kB of PR diffs, 3 kB of issue/PR text, <1 kB of discussion (1 comments), 0 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.