@vtxmacro/cli 2026.8.10 → 2026.8.12
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/README.md +57 -3
- package/bin/vtx.js +5484 -4528
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -39,7 +39,13 @@ vtx-mcp
|
|
|
39
39
|
|
|
40
40
|
The CLI and MCP server use the same saved token and environment variables.
|
|
41
41
|
|
|
42
|
-
##
|
|
42
|
+
## Agent inference hosts
|
|
43
|
+
|
|
44
|
+
Both paths use a separate least-privilege `insights:inference` OAuth grant.
|
|
45
|
+
VTX never receives an agent vendor password, OAuth token, or subscription
|
|
46
|
+
credential.
|
|
47
|
+
|
|
48
|
+
### Automated Codex host
|
|
43
49
|
|
|
44
50
|
```bash
|
|
45
51
|
vtx inference-host login
|
|
@@ -57,13 +63,61 @@ OAuth grant. It also keeps its ChatGPT subscription login in a dedicated private
|
|
|
57
63
|
Codex home. Neither credential is copied from or widens the ordinary CLI or Codex
|
|
58
64
|
session. VTX logout/revoke and Codex logout are intentionally separate operations.
|
|
59
65
|
|
|
66
|
+
### Any compatible agent harness
|
|
67
|
+
|
|
68
|
+
Codex, Claude Code, Cursor, GitHub Copilot, Antigravity, and other agents that can
|
|
69
|
+
run shell commands can use their existing first-party session without a
|
|
70
|
+
VTX-owned vendor adapter:
|
|
71
|
+
|
|
72
|
+
```bash
|
|
73
|
+
vtx inference-host login
|
|
74
|
+
vtx inference-host agent-connect \
|
|
75
|
+
--adapter <lowercase-harness-id> \
|
|
76
|
+
--model <exact-model-id> \
|
|
77
|
+
--model-label "<display label>" \
|
|
78
|
+
--effort <reasoning-effort> \
|
|
79
|
+
--json
|
|
80
|
+
vtx inference-host agent-run
|
|
81
|
+
vtx inference-host agent-next --wait-seconds 50 --json
|
|
82
|
+
```
|
|
83
|
+
|
|
84
|
+
`agent-next` returns separate exact `system_prompt`, `user_prompt`,
|
|
85
|
+
`context_json`, and `output_schema_json` fields. The agent reasons in its own
|
|
86
|
+
session, then writes one JSON object to `agent-complete` stdin. At minimum that
|
|
87
|
+
object contains a `result` string. If the harness exposes exact token usage,
|
|
88
|
+
include the published usage object; otherwise VTX records usage as explicitly
|
|
89
|
+
unavailable. For `agent-fail`, choose the truthful dispatch state and set
|
|
90
|
+
`retryable` true only when a fresh attempt is safe:
|
|
91
|
+
|
|
92
|
+
```json
|
|
93
|
+
{"dispatch_outcome":"not_dispatched","failure_category":"adapter","failure_code":"harness_unavailable","retryable":true}
|
|
94
|
+
{"dispatch_outcome":"confirmed_dispatched","failure_category":"adapter","failure_code":"no_reliable_result","retryable":false}
|
|
95
|
+
{"dispatch_outcome":"outcome_unknown","failure_category":"transport","failure_code":"dispatch_confirmation_lost","retryable":false}
|
|
96
|
+
```
|
|
97
|
+
|
|
98
|
+
Use `not_dispatched` only when no harness inference began,
|
|
99
|
+
`confirmed_dispatched` when it began but returned no trustworthy result, and
|
|
100
|
+
`outcome_unknown` when dispatch itself cannot be confirmed. Pipe exactly one
|
|
101
|
+
object to `vtx inference-host agent-fail --json`.
|
|
102
|
+
|
|
103
|
+
Keep `agent-run` open in a separate terminal and keep calling `agent-next`
|
|
104
|
+
while the host should remain available. The independent keeper preserves
|
|
105
|
+
truthful liveness while the harness reasons. This same loop
|
|
106
|
+
supports Main, Review, and Screener; Provider response and Decision candidate;
|
|
107
|
+
and normally running VTX Client Mode or Server Mode bots. Server Mode keeps the
|
|
108
|
+
normal VTX call fee. Client Mode has no VTX platform fee. Neither mode falls back
|
|
109
|
+
to another model when the agent host is unavailable.
|
|
110
|
+
|
|
60
111
|
## Remove
|
|
61
112
|
|
|
62
|
-
First
|
|
63
|
-
uninstall the package
|
|
113
|
+
First stop the foreground run or agent-run process, revoke the VTX inference
|
|
114
|
+
grant, then uninstall the package. Cleanup refuses while a foreground host
|
|
115
|
+
process still owns the host. Automated Codex users should also remove the
|
|
116
|
+
dedicated Codex login:
|
|
64
117
|
|
|
65
118
|
```bash
|
|
66
119
|
vtx inference-host revoke
|
|
120
|
+
# Automated Codex host only:
|
|
67
121
|
vtx inference-host codex-logout
|
|
68
122
|
npm uninstall -g @vtxmacro/cli
|
|
69
123
|
```
|