fix: migrate goals across compression sessions

[kortylokai-web]

What does this PR do?

Fixes /goal state drift across context-compression session splits.

Hermes can end an oversized parent session and continue in a child session after context compression. Before this change, active /goal state could remain attached to the old parent session while the live conversation continued under the child session. That made post-turn goal continuation/status desynchronize from the active chat, and some paths could revive stale terminal goal rows.

This PR fixes the migration at both important seams:

• the core compression boundary in agent/conversation_compression.py, so CLI/gateway/other callers of _compress_context() inherit the migration;
• gateway session-routing in gateway/run.py, so session-store persistence, Telegram topic binding repair, and /goal state migration stay aligned when a verified compression relationship changes the active session id.

Goal migration is intentionally narrow:

• only verified compression continuations migrate;
• active/paused goals move to the child session;
• the old goal row is preserved as cleared/non-resumable;
• missing, done, cleared, and non-compression child relationships are no-ops;
• /goal resume cannot revive migrated/cleared terminal rows.

Related Issue

N/A — bug found during local gateway/session-routing work.

Existing overlap checked

I searched existing open PRs before and after opening this review surface. This overlaps with the broader goal-compression area, especially:

• fix(cli, gateway): migrate /goal across all session_id rebind sites #18749 — fix(cli, gateway): migrate /goal across all session_id rebind sites
• fix: preserve goals across compression session splits #18427 — fix: preserve goals across compression session splits
• fix(agent): migrate goal state across compression session rotation #34035 — fix(agent): migrate goal state across compression session rotation
• fix: preserve goals across compression sessions #38231 — fix: preserve goals across compression sessions

Compared with those, this PR combines a core compression-boundary hook with gateway-specific routing/topic/session-store guardrails. It includes regression coverage for compression-only migration, non-compression child no-leak behavior, Telegram topic compression-tip repair, manual /compress, hygiene compression, and a guard against future direct session_entry.session_id mutation outside the shared gateway helper.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

Changes Made

• agent/conversation_compression.py
  • migrates active/paused /goal state at the core compression session split after the child session is created.
• hermes_state.py
  • adds atomic state-meta compare/move support;
  • adds verified compression-chain walking helpers.
• hermes_cli/goals.py
  • adds compression-validated goal migration;
  • prevents /goal resume from reviving cleared/done/migrated goals.
• cli.py
  • adds a manual /compress fallback migration call at the CLI session-id rebind seam.
• gateway/run.py
  • routes agent-result compression, Telegram topic compression-tip repair, hygiene compression, manual /compress, and late session-split detection through a shared compression session-switch helper;
  • keeps session-store persistence, Telegram topic binding, and /goal migration aligned.
• Tests
  • adds core compression-boundary regression coverage;
  • adds CLI manual /compress fallback coverage;
  • adds gateway regressions for goal migration across compression splits/tip-walks;
  • covers non-compression child no-leak behavior;
  • covers hygiene compression and manual gateway /compress;
  • updates the session-id persistence guard so future direct session-id mutation outside the helper is caught.

How to Test

Focused verification on Android 16 / Termux after rebasing onto current upstream/main:

PY="$HOME/.hermes/hermes-agent/venv/bin/python"
PATH="$PATH:/bin" "$PY" -m pytest \
  tests/run_agent/test_compression_boundary_hook.py \
  tests/cli/test_manual_compress.py \
  tests/hermes_cli/test_goals.py \
  tests/gateway/test_goal_compression_split.py \
  tests/gateway/test_compression_session_id_persistence.py \
  tests/gateway/test_telegram_topic_mode.py::test_managed_topic_binding_reuses_restored_session_over_static_lane_session \
  tests/gateway/test_telegram_topic_mode.py::test_topic_binding_follows_compression_tip_on_read \
  tests/gateway/test_compress_plugin_engine.py \
  tests/test_hermes_state.py::TestStateMeta \
  -q -o 'addopts=' --tb=short

Result:

90 passed in 38.20s

Additional local checks:

git grep -n '<<<<<<<\|>>>>>>>\|=======' -- <changed files> || true
git diff --check upstream/main...HEAD
"$PY" -m py_compile <changed Python files and focused tests>

Result: no conflict markers in changed files, whitespace check passed, and py_compile passed.

Full-suite status:

The full-suite checkbox remains unchecked. On this Android/Termux phone, a clean full-suite proof currently depends on the separate Termux/test-portability PRs. Without those companion changes, a full-suite run on this phone cannot honestly be marked green for this branch alone.

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: Android 16 / Termux

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 or workflows — or N/A
• I've considered cross-platform impact (Windows, macOS) per the compatibility guide — or N/A
• I've updated tool descriptions/schemas if I changed tool behavior — or N/A

Screenshots / Logs

N/A — behavior covered by regression tests.

[teknium1]

Thanks for this — the diagnosis and fix are correct for the rotation path.

Context on why we're closing it: as of #52658 (#38763), compression.in_place now defaults to True. Compaction keeps ONE durable session id for the conversation's whole life — it no longer ends the session and forks a continuation id. Because the id no longer rotates, the active /goal stays attached to the same session automatically — there is no rotation boundary to migrate the goal across.

Rotation now only runs as an explicit opt-out (compression.in_place: false), and the direction is to treat that path as legacy. So this fix hardens a code path that no longer executes by default and that we're not investing further in. Closing as superseded by the in-place default — not a reflection on the code, which was a correct fix for the behavior at the time. Credit preserved in the issue/PR history. Appreciate the contribution.