Ha Self UpdateCAUTION
🦭 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
Overview
🦭 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
f50447283b3fOBSERVED · 2026-10-12What it tells the agent
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-update
description: "Check for and install Hope Agent updates through conversation. Use whenever the user asks about upgrades, new versions, release notes, or reports a bug that might already be fixed upstream — phrases like 'upgrade Hope Agent', 'update hope agent', 'check for new version', '升级一下', '有新版本吗', '帮我升级', 'is there a newer build', 'check release notes', 'install the latest'. Also use proactively when an `app_update(action=\"check\")` result shows `has_update: true` and the user hasn't been told yet. Covers all three formfactors: desktop GUI bundle (DMG/MSI/AppImage), `hope-agent server` daemon installed via Homebrew/Scoop/AUR/apt/dnf, and headless single-binary deployments. The upgrade is always user-confirmed via `ask_user_question` — never silent."
always: false
aliases:
- self-update
- upgrade
---
# Hope Agent Self-Update
Hope Agent ships a single binary (`hope-agent`) that dispatches into three modes by subcommand: desktop GUI, `hope-agent server`, and `hope-agent acp`. All three share the same release artifacts under [github.com/shiwenwen/hope-agent/releases](https://github.com/shiwenwen/hope-agent/releases) and the same Minisign-signed update manifest. The `app_update` tool is the single entry point for self-upgrade; this skill is the methodology for using it well.
## When to suggest an upgrade
Trigger paths (any one is enough):
- User asks ("升级一下" / "is there an update" / "show release notes" / "the latest version").
- User reports a bug. Run `app_update(action="check")` first — if `has_update: true`, mention "there's a newer version `latest_version`; let me check its notes for `<bug-description>` before we dig in" and read the notes. If the notes mention the bug, suggest upgrading first.
- A startup snapshot shows the agent hasn't upgraded in a long while AND `has_update` is true. Don't nag — bring it up once per conversation at most.
Do NOT trigger if the user is mid-task on something that would be disrupted (active chat turn from a subagent, in-flight cron run, IM media uploading). Wait or finish first.
## Workflow
### 1. Check — never skip
```
app_update(action="check")
```
Returns:
```json
{
"current_version": "0.1.1",
"latest_version": "0.2.0",
"has_update": true,
"install_source": { "kind": "brew", "prefix": "/opt/homebrew" },
"recommended_path": "package_manager",
"platform_target": "darwin-aarch64",
"notes": "fix: ...",
"pub_date": "2026-05-12T10:00:00Z",
"bare_binary_available": true
}
```
Read the result aloud (versions + 1-2 lines from notes). If `has_update == false` and the user asked to upgrade anyway (e.g. for a force-reinstall), surface that and ask if they really want to pin a `target_version`.
### 2. Pre-flight before recommending install
Quick checks before calling `app_update(action="install")` — these aren't blocking, but mention them:
- Use `exec` to peek at active workload: `launchctl list | grep hopeagent` / `systemctl --user is-active hope-agent.service`, plus `sessions_list({ limit: 5 })` to see if any session is mid-turn.
- If a cron job is scheduled in the next few minutes, suggest waiting.
- On macOS: warn if `Hope Agent.app` is open and the user is mid-task — restart will kill in-flight turns.
### 3. Install — pass `run_in_background: true`
```
app_update(action="install", run_in_background: true)
```
`run_in_background: true` is recommended for `install` (download + verify + swap takes 10s-2min depending on connection). The tool returns `{ job_id, status: "started" }` immediately. The user sees a confirmation dialog (Yes/No) — that dialog cannot be bypassed.
If they decline, the tool returns `cancelled_by_user`. Don't try again automatically — wait for the user to ask.
### 4. Track progress
Two ways to follow along:
- `app_update(action="status", job_id="...")` — polls the in-memory phase tracker.
- Frontend subscribes to EventBus topic `app_update:progress` — the UI renders the progress bar automatically; the tool also emits `app_update:completed` when the job finishes.
Phases (in order): `starting → running → downloading → verifying → staging → backing → swapping → restarting → done`. Failure transitions straight to `failed` with an `error` field.
### 5. Verify after install
The service restarts itself on success, but the binary swap is only visible to processes started AFTER the swap — the model conversation is still running the old image. Tell the user:
> Upgrade succeeded. The server service has been restarted on the new image. Your desktop GUI is still running the old version — quit and reopen to load v0.2.0.
Then `exec` to confirm:
```bash
hope-agent --version # should print the new version
curl -s http://127.0.0.1:8420/api/health # if server is configured
```
## Path routing
`recommended_path` in the check output is the auto-selected route. The user can override via `prefer_path` on `install`:
| Path | When | What happens |
| ----------------- | --------------------------------------------- | ---------------------------------------------------------------------------- |
| `tauri` | Desktop GUI in foreground, bridge registered | `tauri-plugin-updater` downloads + verifies + installs the signed bundle. |
| `package_manager` | brew / scoop / apt / dnf / AUR install | Runs `brew upgrade --cask hope-agent` (etc.), then restarts the service. |
| `self_contained` | Manual install, or above paths unavailable | Downloads bare-binary tar.gz, verifies Minisign sig, atomic-swap, restart. |
| `manual_prompt` | Cannot pick automatically | Tool prompts the user via `ask_user_question` to choose recovery. |
`prefer_path: "self_contained"` is the right fallback when the package-manager path fails (stale tap, sudo refused, etc.). `prefer_path: "package_manager"` is rarely needed — only when thTrust audit
CAUTIONgrade C · trust 79/100 Install with care. The audit found things worth knowing before you trust its output.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | NA |
| L2 | Instruction surface (what it tells the agent) | FAIL |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- none-observed
- Shell
- none-observed
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (2)
| `minisign verify failed` | Re-run install (download may have been truncated). If it persists twice, treat as a release-signing problem — DO NOT bypass; tell the user to report it. |
curl -s http://127.0.0.1:8420/api/health # if server is configured
Gates applied: instruction_override, no_behavioural_pass.
f50447283b3ffull audit observations/trust-audit/skill/shiwenwen__ha-self-update.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-12 | f50447283b3f | CAUTION | C | 79 | first audit |
Questions
What does the Ha Self Update skill do?
🦭 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
Is Ha Self Update safe to install?
With care. The audit graded it C (79/100) and found 2 things worth knowing before you trust this skill, listed below with the exact line each was found on.
What can Ha Self Update access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
How current is this page?
The grade is for one exact copy of the source (f50447283b3f), read on 2026-10-12. The repository is watched, and a new audit runs when it changes — this is the first audit.