Hermes: discover and deliver into the current active Codex turn

[100yenadmin]

Summary

Hermes can use the published LCO 1.6.1 MCP surface for indexed-session recall
and approval-gated control of an idle disposable Codex task, but it cannot yet
reliably discover and deliver into the Codex task whose turn is currently
running.

This is distinct from #795. The codex_thread: normalization fix is live and
passed an exact-candidate live-send canary against an idle disposable task.

Verified runtime evidence

• Eva's isolated package is 1.6.1.
• The active Hermes MCP connection succeeds and discovers 76 configured tools.
• The strict Hermes smoke passes the 14 required first-class tools.
• The exact 1.6.1 candidate completed one approval-gated live send using a
  public codex_thread: reference against an idle disposable task.
• In a direct Hermes acceptance run, target selection and dry-run approval
  completed, but live lco_codex_send_message returned
  LIVE_SEND_INTERNAL_ERROR.
• A follow-up read-only run could not find the current executing release task
  in lco_recent_sessions, reported stale/low-confidence inventory, and had no
  active-turn identifier or configured active-thread-state read tool.

Only public-safe aggregate status is retained. No thread identifiers, turn
identifiers, queries, transcripts, paths, credentials, or runtime
configuration are included.

Product gap

lco_codex_send_message starts a new turn and is appropriate for an idle
thread. A running task needs turn/steer, but the current Hermes-first facade
does not provide a reliable current-task discovery plus active-turn handle
flow. The one-shot stdio boundary also cannot be assumed to own the same
active app-server runtime.

Acceptance criteria

• Add a public-safe read path that resolves the current active Codex task
  from live app-server state without requiring session reindexing.
• Expose an opaque, approval-bindable active-turn reference without
  leaking raw thread or turn identifiers.
• Add one Hermes-first delivery operation that deterministically chooses
  new-turn send for idle tasks and safe steer for running tasks, or returns
  a specific public-safe blocker before live mutation.
• Preserve the fixed never-approve/read-only posture and existing dry-run
  audit/hash matching for the exact chosen action.
• Prove bidirectional Hermes to LCO to active Codex delivery in an isolated
  canary with no Telegram involvement.
• Keep all shared evidence free of raw identifiers, prompts, transcripts,
  paths, credentials, and runtime configuration.
• Cover idle send, running steer, stale active-turn ref, ambiguous target,
  and no-current-target behavior.

Dependency and scope

Likely depends on or should be delivered with #787's managed local app-server
daemon transport. This belongs to the deeper native-Hermes/runtime lane, not a
retroactive blocker on the shipped 1.6.1 prefixed-reference hotfix.

Non-goals: Telegram routing, Remote Control enablement, GUI automation,
unattended control, Hermes profile changes, or database refresh as an implicit
side effect.

[coderabbitai]

🔗 Related PRs

#372 - feat: add deterministic Codex autonomy tick [merged]
#390 - Fix same-connection Codex live control [merged]
#760 - Preserve safe Codex live-control posture [merged]
#767 - Prepare LCO 1.6.0 release truth [merged]
#792 - fix: stabilize LCO 1.6 for Hermes-first MCP [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[evaos-code-review-bot]

evaOS issue enrichment

Issue: #797 - Hermes: discover and deliver into the current active Codex turn
State: open; milestone: LCO 1.7 — Eva Remote Codex Control

Related issues/PRs: #795, #787.
Existing labels: enhancement, adapter, safety, codex, mcp, orchestrator.
Suggested labels: bug.
Suggested owners: none.

Related context

• 1.6.1: normalize public Codex thread refs before live control #795 - mentioned in issue metadata; inspect for dependency, duplicate, or prior-decision risk.
• Feature: support Codex app-server daemon UDS transport for resumable local control #787 - mentioned in issue metadata; inspect for dependency, duplicate, or prior-decision risk.

Agent-start packet

Problem shape: Hermes: discover and deliver into the current active Codex turn - ## Summary Hermes can use the published LCO 1.6.1 MCP surface for indexed-session recall and approval-gated control of an idle disposable Codex task, but it cannot yet reliably discover and deliver into the Codex task
Product fit: Product/UX-sensitive issue; review against repo vision and user flow quality, not only technical P0/P1 risk.
Implementation wedge: Ship the smallest user-visible path or prototype that can be smoke-tested against the intended experience.

Build / borrow / buy scan:

• Build: identify the smallest implementation wedge that proves the issue without broad rollout.
• Borrow: inspect same-repo prior art, repo memory, GitNexus context, and allowlisted cross-repo examples first.
• Buy/use: run capped external OSS/library/API/current-market research with citations, observed dates, and why each source matters.
• Library choice: compare maintenance, API fit, license, bundle/runtime cost, and integration proof.

Candidate sources:

• VISION.md / repo policy / repo-memory.md / GitNexus packet.
• Same-repo issues and PRs referenced by issue metadata.
• Allowlisted cross-repo GitHub examples with direct issue/PR/repo URLs.
• External OSS repositories or library/API docs with freshness dates.
• Current market/product examples, capped and cited, only when they change implementation direction.

Acceptance criteria:

• Preserve and execute the issue's stated acceptance criteria.
• Include source citations for any external OSS/library/current-market claims.
• Keep issue-enrichment comments sticky and idempotent.
• Keep labels, owners, reviewers, and roadmap fields suggestion-only.

Proof plan:

• Focused unit/fixture test for the changed planner behavior.
• Dry-run issue-enrichment evidence packet before live comment posting.
• Redaction check for issue body, citations, and generated markdown.
• UX/product smoke or screenshot proof when implementation changes user flow.
• Contract or degraded-mode proof before runtime promotion.
• Dependency/license/API-fit note before adding or relying on a package.

Known traps:

• Do not treat related links as proof by themselves; each source needs a reason why it matters.
• Do not scan old issue backlogs or exceed repo-level enrichment throttles.
• External research must stay capped, cited, fresh-dated, and redacted.
• Technical severity alone can miss product/UX regressions; include product-manager review framing.

Non-goals:

• Do not auto-apply labels, owners, reviewers, roadmap fields, or milestones.
• Do not bulk-enrich old backlog issues from this planner output.
• Do not claim external market or OSS research was performed unless cited sources are present.

Context-source taxonomy:

• [enabled] vision_repo_policy_memory_gitnexus: Use repo vision, policy, memory, and GitNexus first before external research.
• [enabled] same_repo_issues_prs: Issue metadata contains same-repo references.
• [enabled] allowlisted_cross_repo_github: Triggered by issue class; keep to configured allowlist and source caps.
• [enabled] external_oss_examples: Use for build/borrow/buy precedent with citations and observed dates.
• [enabled] library_api_docs: Use when implementation may depend on a module, SDK, API, or framework choice.
• [enabled] current_market_examples: Use for product/UX/market positioning issues with freshness dates.
• [enabled] internal_mcp_docs: Use internal docs/MCP context when available; missing context is degraded, not blocking.

Validation suggestions

• Confirm owner, acceptance criteria, and validation evidence before implementation.

Triage gaps

• Owner or reviewer signal not detected in issue body.

No labels, owners, reviewers, or roadmap fields were changed by this bot.

[100yenadmin]

Milestone 15 scope is now the focused Eva remote-control release tracked by #798. This issue covers daemon-owned task discovery and delivery. If the target is owned by Codex Desktop or ownership cannot be proven, LCO must return desktop_observation_required or another public-safe blocker; it must not claim the managed daemon owns the active Desktop turn. Desktop operation stays in Hermes Computer Use and is accepted through #799.