fix(kernel): resolve team/delegate model pins through the config accessor

[senamakel]

Summary

• Resolves the remaining review threads from feat(kernel): host capability adapters, config seam, and the Phase 2 transcript soak #5396 that were still open when it merged.
• Team model pins now resolve through TeamModelConfig::model_for_role — a team pinning only one tier applied it to that tier only, leaving the other on the global default instead of the team's model.
• A blank model pin no longer displaces a working default — TOML accepts model = "", and both the team and delegate paths assigned it unchecked.
• Extends turn-config test coverage to the two fields no test could have caught being dropped.
• Doc fixes: the kill-switch value list, the MD040/MD001 markdownlint violations, and the Phase 4 plan section that described a seam feat(kernel): host capability adapters, config seam, and the Phase 2 transcript soak #5396 actually landed.

Problem

#5396 merged with 18 unresolved threads. Four were real defects in the config seam; the rest were documentation drift and lint.

The two model-pin bugs share a shape: the mapper copied raw Option<String> fields instead of going through the accessor that owns the semantics.

• TeamModelConfig::model_for_role trims, drops empty strings, and falls back across the lead_model/agent_model pair — its own doc says "callers fall back across the pair so configs can specify only one tier without breaking routing". apply_team_models did none of that, so a team pinning only lead_model ran its subagents on the global default rather than the model the user chose for that team.
• DelegateAgentConfig::model is a bare String. apply_delegate assigned it unchecked, so model = "" replaced a working session model with the empty string and failed at dispatch rather than at config load.

The coverage gap was pointed out with unusual precision and is worth restating: compact_context and tool_result_budget_bytes were mapped but unfalsifiable. default_config_maps_to_the_crate_defaults compares against TurnConfig::default(), so a field left at the crate default still passes; per_section_mappers_agree_with_the_composed_one asserts s.turn == turn_config_from(&c.agent), which compares the mapper against itself. A dropped assignment for either field would have stayed green.

Solution

• apply_team_models resolves both roles through model_for_role(true/false).
• apply_delegate trims and ignores a blank pin, warning rather than silently substituting.
• turn_limits_come_from_the_agent_section sets both previously-uncovered fields away from their defaults, with a comment naming why the other two tests cannot catch them.
• AgentConfig::session_shadow_reads' doc now lists disabled and notes case-insensitivity, matching env_kill_switch_engaged, which accepts 0|false|no|off|disable|disabled. The implementation was right and the config doc was incomplete — the reviewer offered "or remove an unsupported value", but the value is supported.
• plan-agents.md §Phase 4: integration_runtime_config: Option<Config> → runtime_config: Option<Arc<Config>> landed in feat(kernel): host capability adapters, config seam, and the Phase 2 transcript soak #5396, so the plan was describing completed work as a future blocker. Rewritten, and the two crate-side blockers found during that review are now linked.

Behaviour change worth flagging: a_team_pinning_only_one_tier_leaves_the_other_on_the_default asserted the buggy behaviour and has been replaced by a_team_pinning_only_one_tier_applies_it_to_both. If the old semantics were intentional, this is the test to argue about — but TeamModelConfig's own doc says otherwise.

Submission Checklist

• Tests added or updated (happy path + at least one failure / edge case) — a_team_pinning_only_one_tier_applies_it_to_both, a_blank_team_pin_does_not_displace_the_default, a_blank_delegate_model_keeps_the_session_default, plus the extended turn-limits assertions.
• Diff coverage ≥ 80% — deferring to the CI gate; the change is small and test-heavy.
• Coverage matrix updated — N/A: no feature rows added, removed, or renamed.
• All affected feature IDs from the matrix are listed under ## Related — N/A: no matrix rows touched.
• No new external network dependencies introduced — N/A: no network code.
• Manual smoke checklist updated — N/A: no release-cut surface touched.
• Linked issue closed via Closes #NNN — N/A: follow-up to a merged PR, no tracking issue.

Impact

Config mapping only, and the adapters remain unwired (Phase 4 repointing is still blocked), so nothing changes at runtime today. The team-pin fix changes which model a team's subagents would resolve to once the seam goes live.

cargo fmt --check, cargo clippy --lib -D warnings, and the full lib suite (12820 / 0) are clean.

One caveat I would rather state than omit: the first full-suite run in this fresh worktree reported 3 failures whose names I did not capture, and four subsequent runs were clean. It looks like first-run state initialisation rather than these changes — they touch only config mapping and docs — but I could not reproduce it to confirm, so I am not claiming it away.

Related

Follow-up to #5396. Two crate-side blockers surfaced by that review are filed upstream:

• ProgressEvent has no tool-completion milestone, so a host cannot report a tool's outcome tinyagents#88 — ProgressEvent has no tool-completion milestone
• ModelResolveRequest carries no model pin, so a definition's exact model id cannot reach the resolver tinyagents#89 — ModelResolveRequest carries no model pin

Summary by CodeRabbit

• Bug Fixes

  • Improved team model selection with trimming, empty-value filtering, and fallback between configured tiers.
  • Blank delegate model settings no longer override the session’s selected model.
  • Added coverage for model selection and context/tool-result budget scenarios.

• Documentation

  • Clarified configuration planning and model-resolution prerequisites.
  • Documented that setting shadow reads to disabled, as well as other falsy values, disables them.
  • Improved Markdown syntax highlighting for the policy flow diagram.

[greptile-apps]

senamakel has reached the 50-credit limit for trial accounts. To continue receiving code reviews, upgrade your plan.

[coderabbitai]

📝 Walkthrough

Walkthrough

The changes update team and delegate model-pin handling, expand related tests, revise agent migration planning, label a Markdown code fence, and document disabled as a falsy shadow-read value.

Changes

Model pin resolution

Layer / File(s)	Summary
Model pin behavior src/openhuman/agent/tinyagents/config.rs	Team pins now trim values, ignore blanks, and fall back across tiers. Blank delegate overrides preserve the session model and emit a warning.
Model pin validation src/openhuman/agent/tinyagents/config.rs	Tests cover tier fallback, blank pins, compact-context settings, and tool-result budgets.

Agent planning updates

Layer / File(s)	Summary
Agent plan status docs/specs/plan-agents.md	The plan updates heading levels, runtime configuration references, migration status, and upstream adapter blockers.

Documentation clarifications

Layer / File(s)	Summary
Documentation wording docs/specs/kernel.md, src/openhuman/config/schema/agent.rs	The kernel diagram fence is labeled text. The shadow-read documentation includes disabled as a falsy value.

Estimated code review effort: 2 (Simple) | ~10 minutes

Suggested reviewers: al629176

Poem

A rabbit checks each model pin,
Trims the blanks and lets hope in.
Plans grow clear and docs align,
Budgets test the flow just fine.
Hop, hop—cleaner paths now shine!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the primary change: resolving team and delegate model pins through the kernel configuration accessor.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/openhuman/agent/tinyagents/config.rs`:
- Around line 151-162: Add verbose, grep-friendly diagnostics in
src/openhuman/agent/tinyagents/config.rs:151-162 around the model_for_role-based
team-pin resolution, logging the resolution result and whether lead and subagent
pins were applied after fallback without model values; in
src/openhuman/agent/tinyagents/config.rs:173-186, log entry and successful
completion of the non-blank delegate model override, also excluding model
values.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: e51b4e5d-702b-4a2b-9daa-ab9c2e806776

📥 Commits

Reviewing files that changed from the base of the PR and between 4dfa6f2 and fd3eb31.

📒 Files selected for processing (4)

• docs/specs/kernel.md
• docs/specs/plan-agents.md
• src/openhuman/agent/tinyagents/config.rs
• src/openhuman/config/schema/agent.rs

[coderabbitai]

📐 Maintainability & Code Quality | 🟠 Major | ⚡ Quick win

Add diagnostics for the changed model-pin flows.

Both flows change session model state but do not provide complete branch diagnostics.

• src/openhuman/agent/tinyagents/config.rs#L151-L162: log the team-pin resolution result and whether lead and subagent pins were applied after fallback.
• src/openhuman/agent/tinyagents/config.rs#L173-L186: log entry and successful completion of a non-blank delegate model override.

Do not log model values.

As per coding guidelines, “Add verbose, grep-friendly diagnostics for new or changed flows, including entry/exit, branches, external calls, retries, state transitions, and errors; never log secrets or full PII.”

📍 Affects 1 file

• src/openhuman/agent/tinyagents/config.rs#L151-L162 (this comment)
• src/openhuman/agent/tinyagents/config.rs#L173-L186

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/openhuman/agent/tinyagents/config.rs` around lines 151 - 162, Add
verbose, grep-friendly diagnostics in
src/openhuman/agent/tinyagents/config.rs:151-162 around the model_for_role-based
team-pin resolution, logging the resolution result and whether lead and subagent
pins were applied after fallback without model values; in
src/openhuman/agent/tinyagents/config.rs:173-186, log entry and successful
completion of the non-blank delegate model override, also excluding model
values.

Source: Coding guidelines