> ## Documentation Index
> Fetch the complete documentation index at: https://docs.useharmonia.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent skills, tools, prompts, and traces

> Canonical runtime inventory for Harmonia's seven Google ADK specialists.

This page answers which method skill an agent may load, which data tools it may call, where its
focused prompt lives, and which deterministic trace proves what actually ran.

<Warning>Skill guidance is method, never evidence. Search is a tool, not a skill. Memory Bank is advisory context, never authority.</Warning>

## Runtime inventory

| Agent                | ADK name and model                                | Typed contract                                             | Filesystem skill                                                                         | Separate tools and research                                                                                      |
| -------------------- | ------------------------------------------------- | ---------------------------------------------------------- | ---------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- |
| [Nimi](/agents/nimi) | `nimi_analyst`, `gemini-3.5-flash`                | `AnalystInput` → `SourceAnalysis`                          | `nimi-analysis-skills` (seven references)                                                | Isolated `nimi_google_search_agent`; optional isolated `nimi_agent_search_agent` for configured Vertex AI Search |
| [Ryan](/agents/ryan) | `ryan_strategist`, `gemini-3.5-flash`             | `StrategistInput` → `StrategistResult` / `ContentStrategy` | `ryan-strategy-skills` (seven references)                                                | Isolated, request-bound `ryan_google_search_agent` only when current public information is necessary             |
| [Temi](/agents/temi) | `temi_editorial_planner`, `gemini-3.5-flash-lite` | `EditorialPlannerInput` → `EditorialPlan`                  | `temi-editorial-planning-skills` (six references)                                        | Six read-only views over one immutable, digest-bound planning snapshot; no search or core MCP server             |
| [Noni](/agents/noni) | `noni_copywriter`, `gemini-3.5-flash`             | `CopywriterInput` → `ContentDraft`                         | `noni-writing-skills` (ten references)                                                   | `search_verified_publications`; native Google Search through isolated `google_search_agent`                      |
| [Dara](/agents/dara) | `dara_editor`, `gemini-3.5-flash`                 | `EditorialReviewInput` → `EditorialAssessment`             | `dara-editing-skills` (seven references)                                                 | Skill loaders only; no data or research tool                                                                     |
| [Maya](/agents/maya) | `maya_presenter`, `gemini-3.5-flash`              | `UiContext` → `SurfacePlan`                                | None                                                                                     | None; deterministic host hydration only                                                                          |
| [Nova](/agents/nova) | `nova_liaison`, `gemini-3.5-flash`                | `LiaisonInput` → `LiaisonAnswer`                           | `trend-scan`, `signal-watch`, `engagement-insights`, `posting-schedule`, or `job-status` | Six bounded reads for trends, engagement, operator feed, job status, and posting windows                         |

## Prompt and runtime locations

| Agent | Focused prompt                        | Skill or tool runtime                  | Trace enforcement                                                      |
| ----- | ------------------------------------- | -------------------------------------- | ---------------------------------------------------------------------- |
| Nimi  | `agent/harmonia_agent/nimi_prompt.py` | `nimi_skills.py`, `nimi_research.py`   | Independent request-bound skill and research traces                    |
| Ryan  | `agent/harmonia_agent/ryan_prompt.py` | `ryan_skills.py`                       | Skill-first trace and optional isolated-search trace                   |
| Temi  | `agent/harmonia_agent/temi_prompt.py` | `temi_skills.py`                       | Skill-first trace plus exact snapshot identity and digest              |
| Noni  | `agent/harmonia_agent/noni_prompt.py` | `noni_skills.py`                       | Skill-first trace plus exact brief-bound research evidence             |
| Dara  | `agent/harmonia_agent/dara_prompt.py` | `dara_skills.py`                       | Skill-first trace with allow-listed resource paths                     |
| Maya  | `agent/harmonia_agent/maya_prompt.py` | `a2ui_presenter.py`                    | Closed entity-catalog and component/reference validation               |
| Nova  | `agent/harmonia_agent/nova_prompt.py` | `nova_liaison.py`, `skills_runtime.py` | One skill precedes compatible reads; claims cite returned evidence IDs |

## Loading and grounding rules

Nimi, Ryan, Temi, Noni, and Dara load their named project-owned skill exactly once, then load only
approved references relevant to the request. Missing, duplicate, out-of-order, irrelevant, or
unapproved resource loads fail closed. Nova selects exactly one routing skill before a compatible
read tool. Maya has no skill or tool trace.

Search-capable agents never place native `google_search` beside filesystem or function tools in the
same model request. A dedicated ADK search agent returns a typed, request-bound result. Deterministic
code preserves and validates grounding chunks, supports, queries, titles, URLs, and search entry-point
metadata before search evidence IDs may be cited.

## Evidence and authority

Valid evidence comes only from the typed invocation: operator context IDs, Nimi moment or angle IDs,
verified performance IDs, eligible Memory Bank fact IDs with Firestore provenance, approved internal
read-tool evidence IDs, or validated native-search evidence IDs. Skill files, prompts, model memory,
and source-coverage ledgers cannot prove customer, market, trend, performance, or source claims.

No agent may approve, transition workflow state, schedule an external effect, publish, access
credentials, create receipts, verify effects, or grant policy exceptions. See the
[authority matrix](/reference/authority-matrix) and [agent contracts](/reference/agent-contracts).