Ha Self DiagnosisSAFE

🦭 A cross-device desktop AI agent with memory, autonomous goals, dynamic workflows, and headless deployment | 会记忆、能持续推进目标、会动态编排多 Agent 的跨端桌面 AI 助手,也可服务化常驻 NAS / 云端

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: ha-self-diagnosis
description: "Self-understanding and issue reporting for Hope Agent itself. Use when the user asks how Hope Agent works internally, asks about its own source code/docs/runtime behavior, reports a bug/failure/slowness/crash, asks to diagnose logs, or asks to create/submit a GitHub issue for a bug, feature request, or improvement (including when there is no bug). Chinese triggers: 自查, 了解自己, 自我诊断, 排查 Hope Agent, 提交 issue, 需求 issue, 功能改进."
license: MIT
context: fork
effort: high
allowed-tools: [read, grep, find, ls, exec, get_settings, app_update, issue_report, ask_user_question, sessions_list, sessions_search, sessions_history, session_status]
status: active
aliases:
  - self-diagnosis
  - self-study
  - report-issue
  - issue-report
---

# Hope Agent Self-Diagnosis

You are Hope Agent investigating Hope Agent. Your job is to understand the app's own implementation and turn findings or user requests into useful answers or GitHub issues.

## Modes

Choose exactly one primary mode from the user's request.

### self-study

Use when the user asks how Hope Agent works, where something is implemented, what a subsystem does, or how to troubleshoot an area without necessarily reporting a bug.

#### Resolve a source root first

self-study is best grounded in real source plus `docs/architecture/`. Pick the first that applies:

1. **Live checkout** — if the working directory is a Hope Agent source tree (has `crates/ha-core/` and `docs/architecture/`), use it directly.
2. **Bootstrap by version** — a packaged install has no source tree. When the question needs implementation detail (not just runtime behavior), clone the matching release once into a reusable cache:
   - Read the running version: `app_update(action="check")` → `current_version` (e.g. `0.10.0`).
   - Reuse `~/.hope-agent/source-cache/v<version>/` if it already exists and looks complete (has `crates/`).
   - Otherwise shallow-clone that exact tag over anonymous HTTPS (never the SSH remote — packaged users can't auth it):
     ```bash
     git clone --depth 1 --branch v<version> \
       https://github.com/shiwenwen/hope-agent.git \
       ~/.hope-agent/source-cache/v<version>
     ```
   - `docs/architecture/` ships inside the clone, so docs and source arrive together. Per-version dir; old ones are safe to delete.
   - **Read-only**: only `read` / `grep` / `find` the tree. Never `cargo build` / `pnpm install` / run repo scripts.
3. **Bundled fallback** — if there is no network, the clone fails, or the tag does not exist (nightly / self-built), fall back to `references/diagnostic-playbook.md` (its Subsystem Reference names the doc, entry module, database, and log category per subsystem) and say the answer is based on bundled references, not live source.

Always check out the version tag, never `main` — `main` carries unreleased code that won't match the user's binary. Do not bootstrap for issue-report or runtime-only diagnosis (logs / sessions / crash journal already explain "why did it fail"); clone only when self-study truly needs to read the implementation.

#### Workflow

Against whichever source root resolved above:

1. Identify the subsystem and read the closest architecture doc under `docs/architecture/`.
2. Inspect the actual source entry points named by the doc or by `references/diagnostic-playbook.md`.
3. Cross-check runtime/config behavior with `get_settings` when useful.
4. Answer with concrete file/module references, data flow, constraints, and likely debugging entry points.

### issue-report

Use when the user asks to submit an issue, create a feature request, record an improvement, or report a bug. A bug is not required: explicit user requests for requirements or improvements are valid issue-report tasks.

Workflow:

1. Classify `kind` as `bug`, `feature`, or `improvement`.
2. Gather context:
   - Bug: version/platform/run mode, recent errors, related session/tool failures, reproduction steps.
   - Feature: user story, motivation, expected behavior, acceptance criteria.
   - Improvement: current friction, proposed behavior, tradeoffs, acceptance criteria.
3. If Issue Reporting setting `duplicateCheckEnabled` is true, call `issue_report(action="search")` with a concise query.
4. Call `issue_report(action="draft")` and show the draft summary to the user if they have not already explicitly asked to submit.
5. Call `issue_report(action="create")` only when the user asked to submit or after they approve the draft. The tool itself will ask for final confirmation before submitting.

Never bypass the `issue_report(action="create")` confirmation. Never paste raw secrets into the issue body. Evidence can be detailed, but it must be redacted and relevant. If no GitHub token is configured, the tool can still submit through the user's authenticated `gh` CLI.

## Diagnostic Sources

Use these in order:

1. Live source and docs — the working-directory checkout, or a version-matched clone under `~/.hope-agent/source-cache/` (see self-study to bootstrap one).
2. Local runtime data under `~/.hope-agent/`: start with `logs.db` and `sessions.db`, then the subsystem-specific store — `memory.db`, `knowledge/index.db`, `cron.db`, `background_jobs.db`, `local_model_jobs.db`, `recap/recap.db`, `canvas/canvas.db` — and non-DB state like `crash_journal.json` / `config.json`.
3. Settings via `get_settings`.
4. Host/runtime dependency reports: prefer the owner-visible **Settings → About → Toolchain Doctor** report, or run `hope-agent doctor --json` through `exec` when the installed binary supports it. Treat its stable diagnostic codes as read-only evidence; never turn diagnosis into an installer, service start, Docker context switch, daemon mutation, or system upgrade.
5. Bundled references in this skill:
   - `references/diagnostic-playbook.md`
   - `references/issue-template.md`

For SQLite diagnostics, follow the read-only rules from `ha-logs`: use `sqlite3 -readonly` or Python URI `mode=ro`; only run `SELECT` / `.schema` /

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.