backeer-codex 0.11.1 → 0.13.0
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/.codex-plugin/plugin.json +1 -1
- package/README.md +50 -2
- package/dist/opencode.mjs +539 -66
- package/package.json +1 -1
- package/plugin.json +1 -1
- package/scripts/check-package.mjs +4 -1
- package/skills/backeer/SKILL.md +60 -5
- package/src/approval-bridge.mjs +24 -9
- package/src/approval-history.mjs +40 -0
- package/src/command-frames.mjs +50 -0
- package/src/companion.mjs +33 -16
- package/src/gateway.mjs +13 -41
- package/src/install-lock.mjs +43 -0
- package/src/installer.mjs +43 -11
- package/src/opencode.mjs +15 -6
- package/src/version.mjs +1 -0
package/README.md
CHANGED
|
@@ -13,6 +13,24 @@ It provides:
|
|
|
13
13
|
- Safe fallback to the normal Codex permission prompt when Backeer is unavailable or not configured.
|
|
14
14
|
- Saved OpenCode connections and an independent exact-session binding for each chat, automatically tracking its subagent tree over live SSE.
|
|
15
15
|
|
|
16
|
+
## Stability update (0.13.0)
|
|
17
|
+
|
|
18
|
+
The terminal companion releases completed tool payloads on item/turn/thread
|
|
19
|
+
completion. Small exact late-response markers live in a protected temporary ledger
|
|
20
|
+
with a 256-entry RAM cache and are removed on normal shutdown. They do not contain
|
|
21
|
+
tool output, command text or permission parameters.
|
|
22
|
+
|
|
23
|
+
The proxy pauses input/output producers when a receiver is slow, resumes without
|
|
24
|
+
dropping messages, and enforces an explicit 64 MiB per-message budget. Mobile
|
|
25
|
+
control SSE uses a separate 1 MiB frame budget, a ten-second partial-frame deadline
|
|
26
|
+
and a 45-second idle deadline. These limits do not truncate live session output.
|
|
27
|
+
Concurrent Backeer installations serialize through a per-user process lock, so an
|
|
28
|
+
older failed installation cannot roll back a newer successful one.
|
|
29
|
+
|
|
30
|
+
The bundled OpenCode implementation is 0.16.0, including lazy command receipts and
|
|
31
|
+
shared-project hub cleanup. Existing connections, exact request IDs, tree tracking
|
|
32
|
+
and all 25 managed tools retain their protocol behavior.
|
|
33
|
+
|
|
16
34
|
## OpenCode connection per chat
|
|
17
35
|
|
|
18
36
|
### Guided first setup (0.4.0)
|
|
@@ -74,7 +92,7 @@ backeer-codex opencode sessions dev
|
|
|
74
92
|
Then tell Codex in a new chat: **“Connect this chat to dev, session ses_….”**
|
|
75
93
|
The tools validate the project and session, then return a unique binding for
|
|
76
94
|
that chat. Other chats can select different sessions without changing it. Each
|
|
77
|
-
binding has its own
|
|
95
|
+
binding has its own projection/cursors, including when multiple chats follow the same
|
|
78
96
|
session. Commands require that binding; no global default or automatic fallback
|
|
79
97
|
exists. Prompt delivery keeps the existing request-ID deduplication rules.
|
|
80
98
|
|
|
@@ -103,7 +121,7 @@ The same binding/cursor is retained; new instructions and answers still require
|
|
|
103
121
|
fresh project/session verification. Missing replay history is reported as a gap,
|
|
104
122
|
never invented or treated as complete. A gap must not be acknowledged until
|
|
105
123
|
`session_status` succeeds for the same target. Disconnecting one chat does not
|
|
106
|
-
close another chat's
|
|
124
|
+
close another chat's subscription, and active bindings do not consume the quota for
|
|
107
125
|
read-only session listings or diagnosis.
|
|
108
126
|
|
|
109
127
|
The plugin does not create SSH tunnels or change SSH permissions. Keep the saved
|
|
@@ -309,3 +327,33 @@ remain authoritative. Actual native deadlines are reported distinctly as
|
|
|
309
327
|
`native_session_api_timeout` (504), with no automatic command replay.
|
|
310
328
|
Update both ends and load the new remote owner code; a local upgrade alone
|
|
311
329
|
cannot override the remote gateway's old 1.5-second limit.
|
|
330
|
+
|
|
331
|
+
## 0.12.0: durable delivery and economical live control
|
|
332
|
+
|
|
333
|
+
Requires Backeer OpenCode 0.15.0 for all new bridge features. The self-contained
|
|
334
|
+
adapter exposes 25 tools, including connection_report, instruction_status,
|
|
335
|
+
tail_output, read_errors and tool_result. Existing bindings follow verified
|
|
336
|
+
subagents automatically. Ordinary child questions can be read/answered from
|
|
337
|
+
the same chat with an explicit targetSessionId and fresh native ancestry/form
|
|
338
|
+
checks. Instructions continue to target the selected root.
|
|
339
|
+
|
|
340
|
+
Activity uses SSE projection and bounded periodic reconciliation. Pass
|
|
341
|
+
changesOnly:true to receive deltas after the first full card. Critical alerts
|
|
342
|
+
still arrive immediately. Compatible bindings share one project SSE transport
|
|
343
|
+
within one MCP process, retaining independent scopes/cursors and gap recovery.
|
|
344
|
+
Separate MCP processes do not share a transport.
|
|
345
|
+
|
|
346
|
+
Accepted, queued, delivered and completed are distinct. Inspect the persisted
|
|
347
|
+
exact request receipt before retrying uncertain delivery. Protected journals
|
|
348
|
+
store signatures/receipts, never prompt or answer text. Interrupted reservations
|
|
349
|
+
are not replayed automatically. Guarantees begin with requests recorded by the
|
|
350
|
+
new bridge. Native V1 delivery can remain unknown. Session completion is never
|
|
351
|
+
claimed as proof of per-instruction completion.
|
|
352
|
+
|
|
353
|
+
Use `backeer-codex opencode report <name>` or connection_report for a safe,
|
|
354
|
+
layered diagnostic report. Loaded runtime versions are separate from unverified
|
|
355
|
+
installation state. Unsupported protocol majors block commands. Tail/error/tool
|
|
356
|
+
reads return source references and explicit clipping/retention gaps without
|
|
357
|
+
consuming transcript/detail cursors. Metrics measure bytes and latency, not
|
|
358
|
+
billed tokens. Install stages a complete replacement and restores local files
|
|
359
|
+
and marketplace on activation failure; it never edits an active plugin cache.
|