@tuned-tensor/cli 0.9.2 → 0.11.0

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/README.md CHANGED
@@ -63,6 +63,11 @@ Selection metadata can be overridden per process with
63
63
  After configuration, run `tt` to open one terminal for conversation and
64
64
  commands:
65
65
 
66
+ Before the shell opens, `tt` performs a short, non-blocking npm version check.
67
+ If a newer stable release is available it recommends
68
+ `npm install -g @tuned-tensor/cli@latest`; offline or unavailable registry
69
+ checks are ignored.
70
+
66
71
  ```text
67
72
  tt cloud support-agent › What happened in my latest training run?
68
73
  TT Your latest run completed...
@@ -73,7 +78,11 @@ tt cloud support-agent › Ask whether my dataset is ready to train
73
78
  Ordinary sentences go through the locally orchestrated Pi model session. The
74
79
  model can call a
75
80
  strict, bounded set of authenticated Tuned Tensor read tools and prepare-only
76
- mutation tools; it has no shell or filesystem tool. Known CLI commands such as
81
+ mutation tools; it has no shell or general filesystem tool. In `local` mode it
82
+ can prepare one new folder directly beneath the shell's current working
83
+ directory with a validated `tunedtensor.json`. The tool refuses path traversal,
84
+ symlinked workspace roots, existing targets, and unsupported spec fields. It is
85
+ not available to cloud-mode turns. Known CLI commands such as
77
86
  `runs list`, `doctor`, and `models list` still execute directly. Prefix a
78
87
  command with `:` when you want to make that intent explicit. Commands are
79
88
  routed to the mode shown in the prompt; prefix one with `cloud` or `local` to
@@ -92,9 +101,19 @@ Read operations execute immediately. Creating or updating a behaviour spec is
92
101
  a prepare-only model operation: the agent shows the exact proposed action, then
93
102
  waits for `/approve` or `/reject`. Starting or cancelling training remains an
94
103
  explicit `tt runs ...` command outside the model tool loop.
104
+ Local spec creation also waits for `/approve`, never calls the cloud API, requires
105
+ at least two examples for local validation, uses exclusive private folder/file
106
+ creation, and never overwrites an existing path. Secure filesystem mutation is
107
+ currently supported on Linux, where writes are anchored to open workspace and
108
+ destination-directory handles. The proposal is bound to the current workspace
109
+ so changing directories before approval fails safely instead of writing
110
+ somewhere else. Ambiguous write failures remain sealed as `outcome_unknown` for
111
+ manual inspection rather than deleting a path that could belong to a racing
112
+ writer.
95
113
  `/approve` executes deterministic local code with an at-most-once mutation
96
- attempt. Before dispatch, the CLI requires the API to advertise mutation-guard
97
- support; older or incompatible servers are refused. The server conditionally
114
+ attempt. For hosted mutations, before dispatch the CLI requires the API to
115
+ advertise mutation-guard support; older or incompatible servers are refused.
116
+ The server conditionally
98
117
  checks approved spec versions and assigns the action ID as an idempotent create
99
118
  ID. If a response or final state write is lost after dispatch, the action is
100
119
  retained as `outcome_unknown`, cannot be retried automatically, and directs the