feat: expose compression no-op status

[TurgutKural]

Summary

• Expose public last_compression_status, last_compression_noop_reason, and last_compression_was_noop properties.
• Update the fresh-tail preflight regression test to assert through the public API instead of private _last_compression_* fields.

Why

• Hermes host runtimes need a stable public contract to distinguish real LCM compactions from no-op compression attempts.
• The underlying no-op decision already exists in LCM; this PR makes that decision observable without requiring hosts to reach into private implementation state.

Validation

• Focused validation: PYTHONPATH=/path/to/Hermes-Agent python3 -m pytest tests/test_lcm_engine.py::test_lcm_tool_status_includes_optional_cache_usage_metrics tests/test_lcm_engine.py::TestEngineABC::test_preflight_does_not_request_compaction_when_only_fresh_tail_is_over_threshold -q -> 2 passed
• Default validation:
  • pytest tests/test_lcm_core.py tests/test_lcm_engine.py tests/test_packaging_install.py -q
  • pytest -q
  • bash -lc 'ulimit -n 1024 && pytest -q'
  • python -m compileall -q .
  • python -m py_compile scripts/import_lossless_claw.py
  • bash -n scripts/install.sh scripts/update.sh
  • git diff --check -> clean
• Workflow validation, if workflows changed: N/A (no workflow changes)

Notes

• Code intentionally does not change compression behavior; it only exposes existing last-compression status state via public read-only properties.
• Follow-up host-side usage is in fix(compression): skip session split on plugin no-op NousResearch/hermes-agent#58495, with a compatibility fallback for older LCM versions.

Refs #

[stephenschoettler]

Thanks for the PR. Could you update the description to match the repo template before I review?

Please add:

• Validation

The code can stay as-is. This just makes review, release notes, and future archaeology easier.

[Tosko4]

@TurgutKural @stephenschoettler I checked this against current main and did a local validation pass.

My read: this is a legit small contract fix, not placebo. The underlying no-op state already exists and is already set by LCM in the relevant paths (should_compress_preflight() and compress()), especially the high-pressure-but-only-fresh-tail/raw-backlog-below-leaf-threshold case. On current main, there is no public last_compression_* property surface, so a host that needs to distinguish “LCM actually compacted” from “LCM deliberately no-oped” has to either inspect private fields or guess from the returned context. This PR exposes that existing decision as a stable public read surface.

That is in line with the goal of LCM: the engine owns deterministic context-boundary decisions, and the host should be able to observe whether a compaction boundary was real or intentionally skipped without poking private implementation state. It does not change compression behavior or make LCM semantically smarter; it just makes the existing decision inspectable by host/runtime code.

Validation I ran on c82d228ce0896d8e4e7973a232c368c4392f8189:

• python3 -m pytest tests/test_lcm_engine.py::TestEngineABC::test_preflight_does_not_request_compaction_when_only_fresh_tail_is_over_threshold tests/test_lcm_engine.py::test_lcm_tool_status_includes_optional_cache_usage_metrics -q -o addopts= → 2 passed
• python3 -m pytest tests/test_lcm_engine.py -q -o addopts= → 554 passed
• python3 -m pytest -q -o addopts= → 1179 passed, 12 xfailed
• git diff --check origin/main...HEAD → clean
• Synthetic probe: tokens over threshold, preflight returns False, public status is noop, public bool is True, and the public reason includes the leaf-threshold explanation.

One process note: the PR body still uses ## Test; Stephen already asked for the repo-template ## Validation heading, so that should be updated before final review.

[TurgutKural]

@Tosko4 @stephenschoettler updated.

[Tosko4]

@TurgutKural thanks, confirmed the PR description is updated.

One clarification for review: the code head for this PR is still c82d228ce0896d8e4e7973a232c368c4392f8189, so this PR itself remains the LCM-side public contract/plumbing change. It exposes the existing no-op decision through public read-only properties, but it does not change compression behavior by itself.

The actual host-side bug fix for “plugin no-op should not be treated as a successful compression boundary / session split” is the linked Hermes Agent PR: NousResearch/hermes-agent#58495.

So my read is:

• this PR is still valid and useful as the LCM contract surface
• the runtime bug is only fixed when the host-side PR consumes that contract
• the two PRs should be reviewed as related, but not as the same fix

@stephenschoettler flagging that distinction explicitly so the LCM-side review stays scoped correctly.

[stephenschoettler]

Please rebase this onto current main and rerun the focused validation. I want to review this on a current branch before deciding whether it is merge-ready.

[stephenschoettler]

@codex review

Current head: 5fb3d9e9227b.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5fb3d9e922

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Reset stale no-op status before positive preflights

If a session first hits a preflight no-op (for example, only the fresh tail is over threshold) and a later call to should_compress_preflight(...) becomes eligible, this property still reports True because the positive preflight paths return without clearing _last_compression_status or the no-op reason. A host using the new public flag immediately after preflight can therefore misclassify the next real compaction boundary as an LCM no-op until compress() starts. Please clear or update the status/reason on positive preflight decisions before exposing this flag.

Useful? React with 👍 / 👎.

[stephenschoettler]

Approved for merge under Stephen approval: reviewer gate t_9eb57f4e passed final head and verified the stale no-op blocker is fixed; CI is green. The remaining Codex thread is stale/non-material against this head.