@tealbrick/kit 0.3.0-rc.7 → 0.3.0-rc.9

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/BOOTSTRAP.md CHANGED
@@ -1,4 +1,4 @@
1
- > Prerelease 0.3.0-rc.7 targets fresh Eve 0.70 installations. Use the explicit candidate version or `next` tag after publication. Scoped `@tealbrick/*` `latest` remains 0.2.7 for existing Eve 0.66 installations; durable cross-version session migration is not validated.
1
+ > Prerelease 0.3.0-rc.9 targets fresh Eve 0.70 installations. Use the explicit candidate version or `next` tag after publication. Scoped `@tealbrick/*` `latest` remains 0.2.7 for existing Eve 0.66 installations; durable cross-version session migration is not validated.
2
2
 
3
3
  # Deployment guide: Eve agent + Teal Brick kit
4
4
 
@@ -19,7 +19,7 @@ For a fresh machine, the sequence is:
19
19
 
20
20
  1. Install Node 24 and npm.
21
21
  2. Create an Eve 0.70.0 project and configure its primary model credentials.
22
- 3. Install `@tealbrick/kit@0.3.0-rc.7` inside that project.
22
+ 3. Install `@tealbrick/kit@0.3.0-rc.9` inside that project.
23
23
  4. Run `tealbrick setup` to sign in, select capabilities and register the card.
24
24
  5. Build and start Eve, then check a real chat from desktop.
25
25
 
@@ -38,8 +38,8 @@ npm --version
38
38
  ```
39
39
 
40
40
  The first command should report `v24.x`. Run the remaining commands as the account
41
- that will own the new agent project. For our deployment this account runs on
42
- an AVMM guest on Neuu after the September23 migration. This package guide does not provision a VM or install AVMM.
41
+ that will own the new agent project. In a typical deployment this account runs on
42
+ an AVMM guest on the agent host. This package guide does not provision a VM or install AVMM.
43
43
 
44
44
  ## 2. Create a fresh Eve project
45
45
 
@@ -81,7 +81,7 @@ Run this inside the Eve project created in step 2 (or your existing compatible
81
81
  project), not in an empty directory or as a global install:
82
82
 
83
83
  ```bash
84
- npm install --save-exact @tealbrick/kit@0.3.0-rc.7
84
+ npm install --save-exact @tealbrick/kit@0.3.0-rc.9
85
85
  ```
86
86
 
87
87
  This installs Portal, AVM, Voice, Vision, Deliver and the shared provider transport
@@ -184,9 +184,9 @@ does not install AVMM or grant VM administration.
184
184
 
185
185
  For a paired connection use `agentvm-client@<resolvable-hostname>` and the path to
186
186
  that paired user's dedicated SSH key. The host must already be enrolled and its
187
- SSH host key trusted. AVMM binds the key to its user/workspace. For our host, the
188
- hostname is `fedora-n150.taild2144d.ts.net`. Paired mode does not read SSH aliases.
189
- Operator mode can use the configured `nucbox` alias and has that account's authority.
187
+ SSH host key trusted. AVMM binds the key to its user/workspace. For example, the
188
+ hostname might be `agent-host.tailnet.invalid`. Paired mode does not read SSH aliases.
189
+ Operator mode can use a configured SSH alias (for example `agent-host`) and has that account's authority.
190
190
 
191
191
  Named profiles and exact mutation grants can be supplied through a reviewed
192
192
  configuration file using `npx --no-install tealbrick apply config.json`. Follow the
@@ -253,6 +253,5 @@ CRUD, production dispatch and desktop delivery together. Self-evolution requires
253
253
  an existing `deployment/*.ts` native Eve policy module outside the writable agent
254
254
  tree. Native proposals require the deployment's existing governed activation path.
255
255
 
256
- AVMM/guests migrated to Neuu on September23. Verify current canonical migration
257
- records before targeting either host. Do not create new canary agents. Committed
256
+ Verify the deployment's canonical host records before targeting an AVMM host. Do not create new canary agents. Committed
258
257
  source, npm publication and installed deployment are separate states.
package/BUZZ.md ADDED
@@ -0,0 +1,135 @@
1
+ # Run fleet agents from Buzz Desktop on a Mac host
2
+
3
+ Buzz Desktop can create a managed agent and hand it to an external provider.
4
+ The kit ships such a provider: Desktop (on your client Mac) deploys the agent
5
+ over ssh to an agent host Mac, which runs it as `buzz-acp` with
6
+ `tealbrick native acp` as the ACP agent. The kit owner's profile, expert
7
+ contracts and Portal grants stay the authority for what the agent may do.
8
+
9
+ ```
10
+ client Mac: Buzz Desktop -> buzz-backend-<name> --ssh--> host Mac: tb-provider-apply.py -> launchd -> buzz-run.sh -> buzz-acp -> tealbrick native acp
11
+ ```
12
+
13
+ ## Prerequisites
14
+
15
+ - Client Mac: Buzz Desktop, Node 24+, `/usr/bin/python3` (Xcode Command Line Tools).
16
+ - Host Mac: Buzz.app installed (only its `buzz` and `buzz-acp` binaries are used),
17
+ Node 24+, `/usr/bin/python3`, and non-interactive ssh from the client
18
+ (key-based; `ssh agent-host true` must work without a prompt).
19
+ - Host Mac: one kit root per owner under the owners directory (default
20
+ `~/agents/tealbrick-connections/<owner>`), set up with
21
+ `tealbrick native setup` and `tealbrick native enroll`. The host helper refuses
22
+ owners without `<owner>/.tealbrick/native-serve.json`.
23
+
24
+ ## Install
25
+
26
+ On the host Mac:
27
+
28
+ ```sh
29
+ npx -y @tealbrick/kit@next buzz host setup
30
+ ```
31
+
32
+ On the client Mac (where Buzz Desktop runs):
33
+
34
+ ```sh
35
+ npx -y @tealbrick/kit@next buzz provider install
36
+ ```
37
+
38
+ Use the same kit version on both Macs (replace `next` with an exact version to pin it).
39
+ The host setup installs exactly the version it runs as.
40
+
41
+ `buzz host setup` creates `~/agents/buzz/{bin,runtime}` (0700), copies `buzz`
42
+ and `buzz-acp` out of the app bundle into `~/agents/buzz/bin` (so a Buzz.app
43
+ update never changes running agents; re-run setup to pick up a new Buzz), and
44
+ writes `~/agents/buzz/host.json` with the node path, kit CLI, owners directory,
45
+ binary paths, their sha256 and the Buzz.app version. Because npx caches are
46
+ temporary, it installs the same kit version into `~/agents/buzz/runtime` with
47
+ `npm install --prefix` and points the agents at that copy. Pass
48
+ `--kit-cli /absolute/path/to/dist/cli.js` to use an existing install instead.
49
+ Re-running it is safe; it rewrites the host scripts and `host.json`.
50
+
51
+ | Flag | Default |
52
+ |---|---|
53
+ | `--owners-dir DIR` | `~/agents/tealbrick-connections` |
54
+ | `--buzz-app PATH` | `/Applications/Buzz.app` |
55
+ | `--node PATH` | the node running setup |
56
+ | `--kit-cli PATH` | install into `~/agents/buzz/runtime` |
57
+ | `--respond-to-default owner-only\|anyone` | none (Desktop choice applies) |
58
+
59
+ `buzz provider install` writes `~/.local/bin/buzz-backend-tealbrick-mac`
60
+ (0755), which Desktop discovers. `--name NAME` (`[a-z0-9-]{3,40}`) changes the
61
+ provider id, `--bin-dir DIR` the location (it must be on Desktop's PATH or be
62
+ `~/.local/bin`). It refuses to overwrite a file it did not write unless
63
+ `--force` is given.
64
+
65
+ ## Create an agent in Buzz Desktop
66
+
67
+ 1. Restart Buzz Desktop after installing the provider.
68
+ 2. Open **Agents > New agent**.
69
+ 3. Set **Run on** to `tealbrick-mac` (shown as "Teal Brick · tealbrick-mac").
70
+ 4. Set **owner** to the kit owner directory name on the host (for example `owner-a`).
71
+ 5. Set **host** to the ssh alias or `user@host` of the host Mac (for example `agent-host`).
72
+ 6. Create the agent. Desktop deploys it; the deploy waits until the agent subscribes to the relay.
73
+
74
+ The agent key goes only over ssh stdin to the host helper. It is never in a
75
+ command line, an environment of the provider or a log; errors redact keys.
76
+ On the host it is stored as `~/agents/buzz/<label>/nostr.key` (0600) and passed
77
+ only to that agent's `buzz-acp` process. Each agent runs in its own empty
78
+ environment. A failed deploy unloads its launchd unit again.
79
+
80
+ ## Who the agent answers
81
+
82
+ - **Mentions:** every agent gets one rule: it wakes only when its own pubkey is
83
+ p-tagged (an explicit @mention), by anyone. Desktop DMs p-tag the agent, so
84
+ DMs work; untagged channel and thread messages are ignored.
85
+ - **respond_to:** the Desktop choice (`owner-only`, `allowlist`, `anyone`,
86
+ `nobody`) applies. If the host has `~/agents/buzz/fleet-defaults.json`
87
+ (written by `--respond-to-default`), its value replaces only Desktop's
88
+ default `owner-only`; an explicit `allowlist` or `nobody` always wins.
89
+ - Desktop policy settings (session policy, timeouts, parallelism up to 4, and
90
+ similar) pass through an allowlist; anything else is dropped.
91
+
92
+ ## Files and logs on the host
93
+
94
+ | Path | Content |
95
+ |---|---|
96
+ | `~/agents/buzz/host.json` | host paths and versions from setup |
97
+ | `~/agents/buzz/bin/` | `buzz`, `buzz-acp`, `buzz-run.sh`, `buzz-plist.py`, `tb-provider-apply.py` |
98
+ | `~/agents/buzz/<label>/` | one agent: key, auth tag, owner, relay, `acp.env`, `buzz-acp.toml`, `deploy.json` |
99
+ | `~/agents/buzz/<label>/logs/buzz-acp.{out,err}.log` | agent logs |
100
+ | `~/Library/LaunchAgents/com.tealbrick.buzz-acp.<label>.plist` | launchd unit |
101
+
102
+ `<label>` is `<owner>-<first 8 hex of the agent pubkey>-<relay host>-<4 hex>`.
103
+ To stop an agent: `launchctl bootout gui/$(id -u)/com.tealbrick.buzz-acp.<label>`.
104
+
105
+ The agent key and auth tag are read from their files by the last shell
106
+ before `buzz-acp` starts; no process command line ever carries them, only
107
+ their paths.
108
+
109
+ If a precondition is missing (agent directory, owner kit root, relay URL,
110
+ auth tag or key), `buzz-run.sh` writes one `buzz-run: <label>: ...; not
111
+ starting` line to `buzz-acp.err.log` and exits 0. The unit keeps
112
+ `KeepAlive` only for unsuccessful exits, so launchd does not restart it in a
113
+ loop. Fix the cause, then start it again:
114
+ `launchctl kickstart -k gui/$(id -u)/com.tealbrick.buzz-acp.<label>`.
115
+
116
+ ## Remove an agent
117
+
118
+ Buzz Desktop has no undeploy operation for provider-managed agents. Delete
119
+ the agent in Desktop, then on the host Mac:
120
+
121
+ ```sh
122
+ npx -y @tealbrick/kit@next buzz host remove <label>
123
+ ```
124
+
125
+ It boots out the launchd unit, deletes
126
+ `~/Library/LaunchAgents/com.tealbrick.buzz-acp.<label>.plist` and moves
127
+ `~/agents/buzz/<label>` to `~/agents/buzz/.removed/<label>-<timestamp>`
128
+ (0700). The moved directory still holds the agent key; delete it when you no
129
+ longer need it. By hand, the same steps are:
130
+
131
+ ```sh
132
+ launchctl bootout gui/$(id -u)/com.tealbrick.buzz-acp.<label>
133
+ rm ~/Library/LaunchAgents/com.tealbrick.buzz-acp.<label>.plist
134
+ rm -rf ~/agents/buzz/<label>
135
+ ```
package/NATIVE.md CHANGED
@@ -1,10 +1,10 @@
1
1
  # Native agent setup — Stage A candidate
2
2
 
3
- Prerelease suite **0.3.0-rc.7** targets the `next` tag. The seven scoped `@tealbrick/*` packages keep `latest` at 0.2.7, which lacks this flow; the unscoped `tealbrick` facade keeps `latest` at 0.3.0-rc.1. The `tealbrick` facade forwards the existing kit CLI; it is not another SDK. Matching Portal runtime setup endpoints must be deployed before native onboarding can succeed. RC4 added `tealbrick native serve` and `native enroll` (see README.md). RC5 keeps an idle serve process's runtime acknowledged in Portal, and lets the runtime reach Portal-provisioned Knowledge with short-lived Portal grants when no local app binding exists. RC6 adds the owner-only `/tealbrick/v1/capabilities` snapshot (Portal-granted apps and Marketplace consents) used by TBD Chat → Plugins, and `tealbrick native` without a subcommand prints usage. RC7 loads versioned expert contracts (`claude.contracts`) as SDK subagents with scope = contract ceiling ∩ owner grants, pauses approval operations for the owner's payload approval in TBD, and traces every delegation.
3
+ Prerelease suite **0.3.0-rc.9** targets the `next` tag. The seven scoped `@tealbrick/*` packages keep `latest` at 0.2.7, which lacks this flow; the unscoped `tealbrick` facade keeps `latest` at 0.3.0-rc.1. The `tealbrick` facade forwards the existing kit CLI; it is not another SDK. Matching Portal runtime setup endpoints must be deployed before native onboarding can succeed. RC4 added `tealbrick native serve` and `native enroll` (see README.md). RC5 keeps an idle serve process's runtime acknowledged in Portal, and lets the runtime reach Portal-provisioned Knowledge with short-lived Portal grants when no local app binding exists. RC6 adds the owner-only `/tealbrick/v1/capabilities` snapshot (Portal-granted apps and Marketplace consents) used by TBD Chat → Plugins, and `tealbrick native` without a subcommand prints usage. RC7 loads versioned expert contracts (`claude.contracts`) as SDK subagents with scope = contract ceiling ∩ owner grants, pauses approval operations for the owner's payload approval in TBD, and traces every delegation.
4
4
 
5
5
  ## Install and connect
6
6
 
7
- Install with `npm install --save-exact tealbrick@0.3.0-rc.7` (or `tealbrick@next`) once registry publication is verified. Until then, use the review tarballs with the checksum-verifying `install-candidate.mjs`.
7
+ Install with `npm install --save-exact tealbrick@0.3.0-rc.9` (or `tealbrick@next`) once registry publication is verified. Until then, use the review tarballs with the checksum-verifying `install-candidate.mjs`.
8
8
 
9
9
  From your native agent workspace:
10
10
 
@@ -74,7 +74,7 @@ Each request reloads local activation and fetches signed Portal authorization. R
74
74
 
75
75
  ## Acceptance boundary
76
76
 
77
- Disposable installed-artifact tests cover empty-account creation, named selection, permission/ambiguity denial, lost-response retry, signed config, scoped Knowledge-fixture read, revoke/reconnect, deactivation and MCP cancellation. They are not real browser consent or real Knowledge product proof. Root's separate Foxy receipt proves the earlier packed candidate's live Claude MCP inactive denial, exact-session process restart and compaction. Codex and LangChain have independent narrower fixture evidence, not live-provider parity. Marketplace, voice/vision and delegated Tealbrick grants are outside this Stage A claim.
77
+ Disposable installed-artifact tests cover empty-account creation, named selection, permission/ambiguity denial, lost-response retry, signed config, scoped Knowledge-fixture read, revoke/reconnect, deactivation and MCP cancellation. They are not real browser consent or real Knowledge product proof. Root's separate canary-agent receipt proves the earlier packed candidate's live Claude MCP inactive denial, exact-session process restart and compaction. Codex and LangChain have independent narrower fixture evidence, not live-provider parity. Marketplace, voice/vision and delegated Tealbrick grants are outside this Stage A claim.
78
78
 
79
79
  ## Name conflicts and interrupted setup
80
80
 
package/README.md CHANGED
@@ -1,4 +1,4 @@
1
- > Prerelease 0.3.0-rc.7 targets fresh Eve 0.70 installations. Use the explicit candidate version or `next` tag after publication. Scoped `@tealbrick/*` `latest` remains 0.2.7 for existing Eve 0.66 installations; durable cross-version session migration is not validated.
1
+ > Prerelease 0.3.0-rc.9 targets fresh Eve 0.70 installations. Use the explicit candidate version or `next` tag after publication. Scoped `@tealbrick/*` `latest` remains 0.2.7 for existing Eve 0.66 installations; durable cross-version session migration is not validated.
2
2
 
3
3
  # @tealbrick/kit
4
4
 
@@ -9,7 +9,7 @@ availability, not agent permissions. Nothing is mounted by installing the kit.
9
9
  Install inside an existing Eve 0.70.0 project:
10
10
 
11
11
  ```sh
12
- npm install --save-exact @tealbrick/kit@0.3.0-rc.7
12
+ npm install --save-exact @tealbrick/kit@0.3.0-rc.9
13
13
  npx --no-install tealbrick setup
14
14
  ```
15
15
 
@@ -93,6 +93,21 @@ including a single deduplicated provider-transport installation.
93
93
 
94
94
  The `avm` selection accepts the [AVM package connection/policy schema](../avm/README.md): either `{target,sudo}` or `{selectedProfile,profiles}`. Setup can select a saved profile and preserves its grants. New connections can use paired AVMM users or operator SSH. Paired mode requires a dedicated SSH identity path and a resolvable hostname; no credentials are copied. Use `tealbrick apply config.json` for full workspace, ownership, sharing, pool and ingress grants. Disabling AVM removes its entire mount.
95
95
 
96
+ ## Buzz Desktop agents on a Mac host: `tealbrick buzz`
97
+
98
+ Buzz Desktop can deploy a managed agent to another Mac that runs it as
99
+ `buzz-acp` + `tealbrick native acp` for a kit owner. Two commands set it up:
100
+
101
+ ```sh
102
+ npx -y @tealbrick/kit@next buzz host setup # on the agent host Mac
103
+ npx -y @tealbrick/kit@next buzz provider install # on the Mac running Buzz Desktop
104
+ ```
105
+
106
+ Then in Desktop: Agents > New agent > Run on `tealbrick-mac`, with the kit
107
+ `owner` and the ssh `host`. `tealbrick buzz host remove <label>` removes an
108
+ agent from the host. Prerequisites, mention and respond_to rules, flags, log
109
+ locations and removal are in [BUZZ.md](BUZZ.md).
110
+
96
111
  ## Native chat endpoint: `tealbrick native serve`
97
112
 
98
113
  Optional. Lets the owner chat with a native **Claude Agent SDK** or **Codex**
@@ -146,6 +161,8 @@ stops the loop with one `native.runtime.sync.stopped` line. The per-turn
146
161
  hold credentials or delegate. Payload approvals: `approvalOperations`
147
162
  (default `marketplace_execute`) and a contract's outward writes pause the
148
163
  exact call as an `input.requested` approval in TBD; decline or timeout denies.
164
+ Owner **standing grants** (below) can pre-approve bounded classes of these
165
+ calls.
149
166
  Each delegation is traced to `.tealbrick/native-serve/traces/*.jsonl`
150
167
  (owner, expert@version, brief, capabilities used, output digest).
151
168
  - `GET /tealbrick/v1/capabilities` (same owner auth) reports what Portal
@@ -162,6 +179,115 @@ stops the loop with one `native.runtime.sync.stopped` line. The per-turn
162
179
  CA settings, `envPassthrough`) and never `TEALBRICK_*` values. Logs carry ids,
163
180
  tool names and status codes, never prompts, tool inputs or results.
164
181
 
182
+ ### Standing grants (`claude.standingGrants`, opt-in)
183
+
184
+ The owner can pre-approve bounded classes of approval-gated calls, so matching
185
+ calls run without a per-payload prompt. The kit's PreToolUse hook enforces
186
+ them (never the model); a grant only replaces the prompt and never widens
187
+ what Portal grants, the `tealbrickCalls` ceiling or an expert contract allow.
188
+
189
+ ```json
190
+ "standingGrants": [
191
+ {
192
+ "id": "community-telegram",
193
+ "description": "Post event announcements to the community channel",
194
+ "operation": "marketplace_execute",
195
+ "match": {"plugin": "community-telegram", "action": "telegram.send_*"},
196
+ "experts": ["publication-operator"],
197
+ "caps": {"perDay": 10, "perHour": 3, "minIntervalSeconds": 120},
198
+ "expires": "2026-12-31T23:59:59Z",
199
+ "deny": {"inputPattern": ["@everyone", "invoice"]}
200
+ }
201
+ ]
202
+ ```
203
+
204
+ - **Scope.** Up to 64 grants, checked in order, only for
205
+ `mcp__tealbrick__tealbrick_call` calls that would otherwise pause for
206
+ approval (`approvalOperations` or a contract's approval rules). `operation`
207
+ (default `marketplace_execute`) must equal the call's operation. `match`
208
+ needs at least one of `registrationId` (case-sensitive), `plugin`, `action`
209
+ (case-insensitive); each is an exact string or a simple `*` glob, and
210
+ `plugin`/`action` may be lists. Plugin and action come from this runtime's
211
+ own Marketplace grant snapshot for the call's `registrationId`; when they
212
+ cannot be resolved the grant does not match.
213
+ - **Who.** Without `experts` a grant covers the owner session only. `experts`
214
+ lists contract ids whose subagents may also use it (unknown ids are refused
215
+ at load).
216
+ - **Caps.** `perDay` (1-200, rolling 24 hours), optional `perHour` and
217
+ `minIntervalSeconds`. A capped or expired grant falls back to the normal
218
+ payload approval (never a silent deny) and is traced as
219
+ `standing_grant.exhausted`. `deny.inputPattern` entries are case-insensitive
220
+ literal substrings of the call's JSON input; a hit makes the grant not apply.
221
+ - **Expiry.** `expires` (ISO 8601 with offset) is required; expired grants never
222
+ match and are not advertised to the model.
223
+ - **Revoke.** Remove (or edit) the grant in `native-serve.json` and restart
224
+ `native serve` and `native acp`.
225
+ - **Counters.** Usage timestamps per grant id persist in
226
+ `.tealbrick/native-serve/standing-grants.json` (0600, pruned to 24 hours) and
227
+ are shared by `native serve` and `native acp` on the same kit root. Each
228
+ check-and-record runs under an exclusive `standing-grants.json.lock` file
229
+ (`O_EXCL`, retried for up to 3 s; a lock older than 30 s is treated as
230
+ abandoned) and writes are tmp + rename. If the lock or the file is
231
+ unavailable, no standing approval is given.
232
+ - **Audit.** Each use logs `claude.tool` with `approval: "standing:<id>"` and
233
+ traces `owner.approval` with `approval: "standing"`, `grantId`, operation,
234
+ registration, plugin and action (no payload). `GET /tealbrick/v1/capabilities`
235
+ lists every grant with its caps, remaining uses today, expiry and expired
236
+ flag (deny patterns as a count only). The owner session's system prompt
237
+ lists unexpired grants and states that anything else still needs approval,
238
+ and that spending, contracts or signatures and first contact with strangers
239
+ are never covered unless a grant explicitly says so.
240
+ - **`native acp`** has no approval surface: calls covered by a grant proceed,
241
+ everything else is still declined.
242
+
243
+ ### Sandboxed workspace (`claude.workspace`, opt-in)
244
+
245
+ With `"workspace": {"dir": "/abs/root/work", "network": {"allowedDomains": [...]}, "envAllow": [...]}`
246
+ (inside the kit root; never `.tealbrick` or `inbox`), Restricted sessions become
247
+ **Workspace (sandboxed)** sessions:
248
+
249
+ - Tools: `Bash` (always in the Claude Code OS sandbox; `dangerouslyDisableSandbox`
250
+ and background commands are refused), `WebSearch`, `WebFetch`, `Agent` and the
251
+ Teal Brick tools. There is **no** `Read`, `Write`, `Edit`, `Glob`, `Grep`,
252
+ `NotebookEdit` or other in-process file tool: those run outside the sandbox
253
+ (symlink and check/use races), so they are removed, listed in
254
+ `disallowedTools` and denied by the tool gate. The model does all file work
255
+ with shell commands. Experts and contracts get their tools intersected with
256
+ this set, so they lose their file tools in this mode too.
257
+ - `WebFetch` refuses non-http(s) URLs, `localhost`, `*.local`, `*.localhost`,
258
+ `*.internal`, `*.ts.net`, single-label names and loopback, private,
259
+ link-local and CGNAT (100.64/10) IP literals (IPv4 and IPv6).
260
+ - Sandbox filesystem: writes only to the workspace (and its `.tmp`, used as
261
+ `TMPDIR`; git uses `.tmp/gitconfig` as its global config). Reads are denied for
262
+ the SDK process's `HOME`, the owners directory (the kit root's parent),
263
+ `<root>/.tealbrick`, `/Volumes` and `/private/var/folders`; `allowRead`
264
+ re-opens only the workspace, the attachment inbox, the node install prefix,
265
+ Claude Code's `shell-snapshots` and present `/opt/homebrew/{bin,lib,include,share,opt,Cellar}`.
266
+ The SDK documents that `allowRead` takes precedence over `denyRead` for
267
+ matching paths; a unit test pins the generated options.
268
+ - Every `envPassthrough` name is removed from sandboxed commands
269
+ (`credentials.envVars` deny) unless listed in `envAllow`; `BUZZ_*` keys,
270
+ `ANTHROPIC_API_KEY`, `CLAUDE_CODE_OAUTH_TOKEN` and `TEALBRICK_*` are always removed.
271
+ The SDK keeps the host's Claude Code sign-in (`CLAUDE_CONFIG_DIR` is not changed).
272
+ - Buzz attachments (`native acp`) are downloaded by the kit into `<root>/inbox`
273
+ (0700, readable but not writable by the sandbox); the model copies them into
274
+ the workspace with Bash.
275
+ - macOS tools that write to the per-user system temp folder instead of `TMPDIR`
276
+ (for example `sips`) fail in the sandbox, because `/private/var/folders` is
277
+ shared by every owner on the host and stays denied. Install workspace-local
278
+ tools instead (a `.venv` with Pillow, or `npm install sharp`); PyPI and the npm
279
+ registry need to be in `allowedDomains`.
280
+ - Owners on one host run as one OS user. Claude Code's sandbox points Bash `TMPDIR`
281
+ at `/tmp/claude-<uid>`, where every session also keeps its tool output (one
282
+ folder per working directory). The kit makes that root unreadable except this
283
+ owner's own folder, write-denies the other owners' folders (it stays writable
284
+ for Claude Code's own cwd tracking), and rewrites each sandboxed Bash command to
285
+ use the workspace `.tmp` as `TMPDIR`/`TMP`/`TEMP`.
286
+ - No git inside the workspace: Claude Code's sandbox refuses writes to `.git/config`
287
+ and `.git/hooks`, and the kit does not work around it. Version an owner's
288
+ workspace from outside the sandbox (for example a host job that snapshots and
289
+ commits it).
290
+
165
291
  ### Configure
166
292
 
167
293
  `.tealbrick/native-serve.json` (owner-only, `chmod 600`) holds no secrets:
@@ -171,14 +297,14 @@ stops the loop with one `native.runtime.sync.stopped` line. The per-turn
171
297
  "version": 1,
172
298
  "port": 5331,
173
299
  "claude": {
174
- "cwd": "/Users/you/agents/henry-serve/workspace",
175
- "instructionsFile": "/Users/you/agents/henry-calendar/agents/henry/INSTRUCTIONS.md",
176
- "recipe": {"path": "/Users/you/.config/tealbrick-native/config.json", "agent": "henry"},
300
+ "cwd": "/Users/you/agents/agent-serve/workspace",
301
+ "instructionsFile": "/Users/you/agents/agent-profile/agents/agent/INSTRUCTIONS.md",
302
+ "recipe": {"path": "/Users/you/.config/tealbrick-native/config.json", "agent": "agent"},
177
303
  "requireSubscription": true,
178
304
  "defaultTools": ["WebSearch", "WebFetch", "Agent"],
179
305
  "experts": {
180
- "source-researcher": {"promptFile": "/Users/you/agents/henry-calendar/experts/source-researcher.md", "tools": ["WebSearch", "WebFetch"]},
181
- "independent-reviewer": {"promptFile": "/Users/you/agents/henry-calendar/experts/independent-reviewer.md"}
306
+ "source-researcher": {"promptFile": "/Users/you/agents/agent-profile/experts/source-researcher.md", "tools": ["WebSearch", "WebFetch"]},
307
+ "independent-reviewer": {"promptFile": "/Users/you/agents/agent-profile/experts/independent-reviewer.md"}
182
308
  },
183
309
  "nativeTools": ["Read", "Glob", "Grep", "Write", "Edit", "TodoWrite", "WebSearch", "WebFetch"]
184
310
  }
package/RUNTIME.md CHANGED
@@ -26,7 +26,7 @@ The Portal signs a metadata-only configuration lease (maximum five minutes). The
26
26
 
27
27
  ### Canary verification, 2026-09-18
28
28
 
29
- Private candidate tarballs installed only in Aura Canary on Nucbox exposed the three memory reads. The authenticated desktop conversation retrieved a synthetic Knowledge document with its citation; a subsequent scoped fact-only recall returned six facts extracted by native GBrain. Update/Delete, cross-partition reads and agent extraction/admin requests were denied. This is not an npm publication or a general memory-quality certification. Query expansion degraded on one live request; serialized background extraction temporarily blocked entity enumeration, and the completed entity register remained empty despite fact-linked entity slugs. See the LABS report `report/readiness-2026-09-18-knowledge-live-canary.md` for exact deployment and remaining defects.
29
+ Private candidate tarballs installed only in a private canary agent on the deployment host exposed the three memory reads. The authenticated desktop conversation retrieved a synthetic Knowledge document with its citation; a subsequent scoped fact-only recall returned six facts extracted by native GBrain. Update/Delete, cross-partition reads and agent extraction/admin requests were denied. This is not an npm publication or a general memory-quality certification. Query expansion degraded on one live request; serialized background extraction temporarily blocked entity enumeration, and the completed entity register remained empty despite fact-linked entity slugs. See the LABS report `report/readiness-2026-09-18-knowledge-live-canary.md` for exact deployment and remaining defects.
30
30
 
31
31
  ## Native workflows
32
32
 
@@ -0,0 +1,99 @@
1
+ #!/usr/bin/python3
2
+ # tealbrick-buzz-provider: generated by `tealbrick buzz provider install`; safe to overwrite with --force.
3
+ """Buzz Desktop managed-agent provider: runs a Desktop-managed agent on a Mac host as buzz-acp + Teal Brick
4
+ kit `native acp` (the owner's kit profile, expert contracts and Portal grants stay the authority).
5
+ Protocol v1: one JSON request on stdin, one JSON response on stdout. The agent key travels only over
6
+ stdin -> ssh stdin -> host helper stdin; never argv, env or logs."""
7
+ import hashlib, json, os, re, subprocess, sys
8
+ PROVIDER_ID = '__TB_PROVIDER_ID__'; VERSION = '__TB_KIT_VERSION__'
9
+ # Test-only override (a fake ssh); Desktop never sets it. Real deployments always use the system ssh.
10
+ SSH = os.environ.get('TEALBRICK_BUZZ_PROVIDER_TEST_SSH') or '/usr/bin/ssh'
11
+ HELPER = 'agents/buzz/bin/tb-provider-apply.py' # relative to the remote HOME (ssh starts there)
12
+ HEX = re.compile(r'^[0-9a-f]{64}$')
13
+ # --- bech32 (NIP-19 nsec) + secp256k1 x-only pubkey, stdlib only ---------------------------------
14
+ CS = 'qpzry9x8gf2tvdw0s3jn54khce6mua7l'
15
+ def nsec_hex(s):
16
+ s = s.strip().lower()
17
+ if HEX.match(s): return s
18
+ if not s.startswith('nsec1') or any(c not in CS for c in s[5:]): raise ValueError('private key is not nsec/hex')
19
+ data = [CS.index(c) for c in s[5:-6]]; acc = bits = 0; out = []
20
+ for v in data:
21
+ acc = (acc << 5) | v; bits += 5
22
+ while bits >= 8: bits -= 8; out.append((acc >> bits) & 0xff)
23
+ if len(out) != 32: raise ValueError('bad nsec length')
24
+ return bytes(out).hex()
25
+ P = 2**256 - 2**32 - 977; N = 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFEBAAEDCE6AF48A03BBFD25E8CD0364141
26
+ G = (0x79BE667EF9DCBBAC55A06295CE870B07029BFCDB2DCE28D959F2815B16F81798, 0x483ADA7726A3C4655DA4FBFC0E1108A8FD17B448A68554199C47D08FFB10D4B8)
27
+ def add(a, b):
28
+ if a is None: return b
29
+ if b is None: return a
30
+ if a[0] == b[0] and (a[1] + b[1]) % P == 0: return None
31
+ l = (3 * a[0] * a[0] * pow(2 * a[1], -1, P)) % P if a == b else ((b[1] - a[1]) * pow(b[0] - a[0], -1, P)) % P
32
+ x = (l * l - a[0] - b[0]) % P; return (x, (l * (a[0] - x) - a[1]) % P)
33
+ def pubkey(sk_hex):
34
+ k = int(sk_hex, 16)
35
+ if not 0 < k < N: raise ValueError('bad key')
36
+ r, q = None, G
37
+ while k:
38
+ if k & 1: r = add(r, q)
39
+ q = add(q, q); k >>= 1
40
+ return format(r[0], '064x')
41
+ # -------------------------------------------------------------------------------------------------
42
+ CONFIG_SCHEMA = {'type': 'object', 'required': ['owner', 'host'], 'additionalProperties': False, 'properties': {
43
+ 'owner': {'type': 'string', 'title': 'Teal Brick kit owner on the host',
44
+ 'description': 'Kit owner directory name under the host owners directory (set up with tealbrick native setup + native enroll), e.g. owner-a'},
45
+ 'host': {'type': 'string', 'title': 'SSH host',
46
+ 'description': 'SSH host alias or user@host of the agent host Mac (key-based, non-interactive), e.g. agent-host'}}}
47
+ def rules_toml():
48
+ # One rule: an agent wakes only when its own pubkey is p-tagged (an explicit @mention), by anyone.
49
+ # Desktop DMs p-tag the agent, so DMs still work; untagged channel and thread messages are ignored.
50
+ kinds = '[9, 40003, 46010, 40007]'
51
+ return (f'# Generated by buzz-backend-{PROVIDER_ID}: wake only on an explicit @mention (p-tag) from anyone.\n'
52
+ f'[[rules]]\nname = "mentions"\nchannels = "all"\nkinds = {kinds}\nrequire_mention = true\n')
53
+ def deploy(req):
54
+ agent = req.get('agent') or {}; cfg = req.get('provider_config') or {}
55
+ owner = str(cfg.get('owner') or '').strip().lower(); host = str(cfg.get('host') or '').strip()
56
+ if not re.fullmatch(r'[a-z0-9-]{2,32}', owner): raise ValueError('provider_config.owner is required (e.g. owner-a)')
57
+ # A leading letter/digit keeps the host from ever being parsed as an ssh option.
58
+ if not re.fullmatch(r'([A-Za-z0-9][A-Za-z0-9._-]{0,31}@)?[A-Za-z0-9][A-Za-z0-9._-]{0,63}', host):
59
+ raise ValueError('provider_config.host is required (an ssh alias or user@host, e.g. agent-host)')
60
+ sk = nsec_hex(agent.get('private_key_nsec') or ''); apub = pubkey(sk)
61
+ launch = agent.get('launch') or {}; owner_pub = (launch.get('owner_pubkey') or '').lower()
62
+ tag = agent.get('auth_tag')
63
+ if not HEX.match(owner_pub): raise ValueError('payload has no owner_pubkey')
64
+ if not tag: raise ValueError('agent has no NIP-OA auth_tag (recreate it in Desktop)')
65
+ t = json.loads(tag) if isinstance(tag, str) else tag
66
+ if not (isinstance(t, list) and len(t) == 4 and t[0] == 'auth' and t[1] == owner_pub): raise ValueError('auth_tag owner does not match the Desktop owner')
67
+ relay = (agent.get('relay_url') or '').rstrip('/')
68
+ m = re.fullmatch(r'wss?://([A-Za-z0-9-]+)[A-Za-z0-9.-]*(:\d+)?', relay)
69
+ if not m: raise ValueError('payload relay_url is not a ws(s) URL')
70
+ slug = m.group(1).lower()[:20] + '-' + hashlib.sha256(relay.encode()).hexdigest()[:4]
71
+ label = f'{owner}-{apub[:8]}-{slug}'
72
+ remote = {'label': label, 'owner': owner, 'relay_url': relay, 'agent_pubkey': apub, 'owner_pubkey': owner_pub,
73
+ 'private_key': sk, 'auth_tag': json.dumps(t), 'respond_to': agent.get('respond_to'),
74
+ 'respond_to_allowlist': agent.get('respond_to_allowlist') or [], 'name': agent.get('name'),
75
+ 'policy_env': launch.get('policy_env') or {}, 'rules_toml': rules_toml()}
76
+ p = subprocess.run([SSH, '-o', 'BatchMode=yes', '-o', 'ConnectTimeout=15', '--', host, '/usr/bin/python3', HELPER],
77
+ input=json.dumps(remote), capture_output=True, text=True, timeout=180)
78
+ lines = [l for l in p.stdout.splitlines() if l.strip().startswith('{')]
79
+ try: res = json.loads(lines[-1]) if lines else None
80
+ except ValueError: res = None
81
+ if not isinstance(res, dict): res = {'ok': False, 'error': (p.stderr or 'no response from host')[-300:]}
82
+ if not res.get('ok'): raise RuntimeError(f"{host}: {res.get('error')}")
83
+ return {'ok': True, 'agent_id': f'{PROVIDER_ID}:{host}:{label}'}
84
+ def redact(text):
85
+ return re.sub(r'nsec1[0-9a-z]+|[0-9a-fA-F]{64}', '<redacted>', text)
86
+ def main():
87
+ try:
88
+ req = json.loads(sys.stdin.read() or '{}'); op = req.get('op')
89
+ if op == 'info':
90
+ res = {'ok': True, 'name': f'Teal Brick · {PROVIDER_ID}', 'version': VERSION, 'protocol_version': 1,
91
+ 'description': 'Runs the agent on a Mac host over ssh as buzz-acp + Teal Brick kit native acp for a kit owner; Portal grants and kit contracts stay the authority.',
92
+ 'config_schema': CONFIG_SCHEMA}
93
+ elif op == 'deploy': res = deploy(req)
94
+ else: raise ValueError(f'unsupported op {op!r}')
95
+ print(json.dumps(res)); return 0
96
+ except Exception as e:
97
+ msg = redact(f'{type(e).__name__}: {e}')[:400]
98
+ print(json.dumps({'ok': False, 'error': msg})); print(msg, file=sys.stderr); return 1
99
+ if __name__ == '__main__': sys.exit(main())
@@ -0,0 +1,22 @@
1
+ #!/usr/bin/env python3
2
+ # tealbrick-buzz-host: installed by `tealbrick buzz host setup`; rewritten on every setup run.
3
+ """Writes ~/Library/LaunchAgents/com.tealbrick.buzz-acp.<label>.plist for one provider-managed agent.
4
+ The unit runs ~/agents/buzz/bin/buzz-run.sh <label>; logs stay in the agent directory (0700)."""
5
+ import os, plistlib, re, sys
6
+ def main():
7
+ if len(sys.argv) != 2 or not re.fullmatch(r'[a-z0-9][a-z0-9-]{2,62}', sys.argv[1]):
8
+ print('usage: buzz-plist.py <label>', file=sys.stderr); return 2
9
+ label = sys.argv[1]; H = os.path.expanduser('~'); BUZZ = os.path.join(H, 'agents', 'buzz')
10
+ D = os.path.join(BUZZ, label); logs = os.path.join(D, 'logs')
11
+ os.makedirs(logs, mode=0o700, exist_ok=True); os.chmod(logs, 0o700)
12
+ agents = os.path.join(H, 'Library', 'LaunchAgents'); os.makedirs(agents, exist_ok=True)
13
+ unit = f'com.tealbrick.buzz-acp.{label}'
14
+ body = {'Label': unit, 'ProgramArguments': ['/bin/bash', os.path.join(BUZZ, 'bin', 'buzz-run.sh'), label],
15
+ 'WorkingDirectory': D, 'RunAtLoad': True, 'KeepAlive': {'SuccessfulExit': False},
16
+ 'ThrottleInterval': 30, 'ProcessType': 'Background', 'Umask': 63,
17
+ 'StandardOutPath': os.path.join(logs, 'buzz-acp.out.log'), 'StandardErrorPath': os.path.join(logs, 'buzz-acp.err.log')}
18
+ path = os.path.join(agents, unit + '.plist'); tmp = path + '.tmp'
19
+ fd = os.open(tmp, os.O_WRONLY | os.O_CREAT | os.O_TRUNC, 0o600)
20
+ with os.fdopen(fd, 'wb') as f: plistlib.dump(body, f)
21
+ os.chmod(tmp, 0o600); os.replace(tmp, path); print(path); return 0
22
+ if __name__ == '__main__': sys.exit(main())
@@ -0,0 +1,39 @@
1
+ #!/bin/bash
2
+ # tealbrick-buzz-host: installed by `tealbrick buzz host setup`; rewritten on every setup run.
3
+ # Runs one provider-managed agent: buzz-acp with Teal Brick kit `native acp` as its ACP agent.
4
+ # Usage (from its launchd unit): buzz-run.sh <label>
5
+ # Each agent gets an isolated environment (env -i): nothing from the login session leaks into the
6
+ # agent, and one agent's keys never reach another (block/buzz#3142). The key and auth tag are read
7
+ # from their files by the final shell and never appear in any process argv (only their paths do).
8
+ # A missing precondition logs one line and exits 0: the unit uses KeepAlive SuccessfulExit=false, so
9
+ # launchd does not restart-loop an agent that cannot start. Fix the cause, then
10
+ # `launchctl kickstart -k gui/$(id -u)/com.tealbrick.buzz-acp.<label>`.
11
+ set -euo pipefail
12
+ NODE=@@NODE@@
13
+ KIT_CLI=@@KIT_CLI@@
14
+ OWNERS_DIR=@@OWNERS_DIR@@
15
+ BUZZ=@@BUZZ_BIN@@
16
+ BUZZ_ACP=@@BUZZ_ACP_BIN@@
17
+ a=${1:-}
18
+ stop() { echo "buzz-run: ${a:-<no label>}: $1; not starting (exit 0, launchd will not restart it)" >&2; exit 0; }
19
+ [[ "$a" =~ ^[a-z0-9][a-z0-9-]{2,62}$ ]] || stop "bad or missing label"
20
+ D="$HOME/agents/buzz/$a"
21
+ [ -d "$D" ] || stop "agent directory $D is missing"
22
+ O=$(cat "$D/owner" 2>/dev/null || echo "$a")
23
+ [[ "$O" =~ ^[a-z0-9-]{2,32}$ ]] || stop "bad owner in $D/owner"
24
+ R="$OWNERS_DIR/$O"
25
+ [ -f "$R/.tealbrick/native-serve.json" ] || stop "kit owner '$O' is not set up under $OWNERS_DIR"
26
+ RELAY=$(cat "$D/relay-url" 2>/dev/null || cat "$HOME/agents/buzz/relay-url" 2>/dev/null || true)
27
+ [ -n "$RELAY" ] || stop "no relay URL ($D/relay-url)"
28
+ [ -s "$D/auth-tag.json" ] || stop "no auth tag ($D/auth-tag.json)"
29
+ [ -s "$D/nostr.key" ] || stop "no agent key ($D/nostr.key)"
30
+ exec /usr/bin/env -i HOME="$HOME" USER="${USER:-$(id -un)}" LOGNAME="${LOGNAME:-$(id -un)}" \
31
+ PATH="$(dirname "$NODE"):/usr/bin:/bin:/usr/sbin:/sbin" LANG=en_US.UTF-8 \
32
+ BUZZ_RELAY_URL="$RELAY" \
33
+ BUZZ_ACP_AGENT_COMMAND="$NODE" BUZZ_ACP_AGENT_ARGS="$KIT_CLI,native,acp,--root,$R,--buzz-cli,$BUZZ" \
34
+ BUZZ_ACP_RESPOND_TO=owner-only BUZZ_ACP_SESSION_POLICY=thread BUZZ_ACP_PERMISSION_MODE=default \
35
+ BUZZ_ACP_MULTIPLE_EVENT_HANDLING=queue BUZZ_ACP_IDLE_TIMEOUT=1800 BUZZ_ACP_NO_BASE_PROMPT=true \
36
+ BUZZ_ACP_NO_MEMORY=true BUZZ_ACP_CONFIG="$D/buzz-acp.toml" RUST_LOG=info \
37
+ $(grep -E "^BUZZ_ACP_(SUBSCRIBE|NO_MENTION_FILTER|CHANNELS|KINDS|RESPOND_TO|RESPOND_TO_ALLOWLIST|SESSION_POLICY|IDLE_TIMEOUT|MAX_TURN_DURATION|TURN_TIMEOUT|AGENTS|HEARTBEAT_INTERVAL|DEDUP|MULTIPLE_EVENT_HANDLING|RELAY_OBSERVER|LAZY_POOL)=[A-Za-z0-9,._:-]*$" "$D/acp.env" 2>/dev/null || true) \
38
+ /bin/bash -c 'set -eu; BUZZ_PRIVATE_KEY=$(cat "$1"); BUZZ_AUTH_TAG=$(cat "$2"); export BUZZ_PRIVATE_KEY BUZZ_AUTH_TAG; exec "$3"' \
39
+ buzz-run "$D/nostr.key" "$D/auth-tag.json" "$BUZZ_ACP"