You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit cd994a5
Browse filesBrowse the repository at this point in the historyBrowse files
Copy file name to clipboardExpand all lines: EVENTS.md
+8-8Lines changed: 8 additions & 8 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -105,16 +105,16 @@ Emitted once when the session begins.
105
105
106
106
### `tool_catalog`
107
107
108
-
Emitted once per session, immediately after `session_start` and before the first model turn (present even for zero-turn or immediately-failing runs). Lists every tool and skill that was resolved for the session so consumers can structurally verify tool exposure without string-matching the model's prose.
108
+
Emitted once per session, immediately after `session_start` and before the first model turn (present even for zero-turn or immediately-failing runs). Reports builtin registry and resolved MCP tool inventory, plus in-scope skills. This session-level snapshot supports inventory diagnostics; it is not the effective tool list for each model request.
@@ -125,13 +125,13 @@ Emitted once per session, immediately after `session_start` and before the first
125
125
}
126
126
```
127
127
128
-
#### `tools[]` — builtin and MCP tools resolved for this session
128
+
#### `tools[]` — builtin registry and resolved MCP inventory for this session
129
129
130
-
-**`name`** (string, **required**) — the tool/function name as exposed to the model. Mirrors upstream `ToolListItem.id`. For MCP tools this is `{serverName}_{toolName}` (both sanitised to `[a-zA-Z0-9_-]`).
130
+
-**`name`** (string, **required**) — the tool/function identifier used when a tool is dispatched. Mirrors upstream `ToolListItem.id`. For MCP tools this is `{serverName}_{toolName}` (both sanitised to `[a-zA-Z0-9_-]`). The identifier does not guarantee inclusion in a model request.
131
131
-**`source`** (string, **required**) — `"builtin"` for tools built into the CLI; `"mcp"` for tools provided by a connected MCP server.
132
132
-**`server`** (string, optional) — the MCP server (client name from config) that provides this tool. Present only when `source` is `"mcp"`.
133
133
134
-
`description` and `parameters` are intentionally omitted to keep the event lean. The primary consumer only needs `name` + `source`to verify tool exposure.
134
+
`description` and `parameters` are intentionally omitted to keep the event lean. `name` + `source`identify inventory entries and their provenance; they do not prove dispatch exposure.
135
135
136
136
#### `skills[]` — skill packs resolved for this session (separate from `tools[]`)
137
137
@@ -140,9 +140,9 @@ Only skills in scope for the run are listed; see [Skill Events](#skill-events).
140
140
-**`name`** (string, **required**) — skill name from `SKILL.md` frontmatter.
141
141
-**`version`** (string or null, **required**) — version string from the `version` field in `SKILL.md` frontmatter. `null` when no version is declared.
142
142
143
-
> **Use case:**The server-side completion gate (aictrl-dev/aictrl #3216) uses `tools[]` to verify that `record_finding`and `record_review_completed` were actually in the model's function list at dispatch time, and `skills[]` to record which skill version produced the review. Detects the "silent success" failure mode where a missing MCP server lets a run complete green without ever having the review tools available.
143
+
> **Use case:**A server-side completion gate (aictrl-dev/aictrl #3216) can use `tools[]` to detect missing expected inventory and `skills[]` to record resolved skill versions. To verify that `aictrl_record_finding`and `aictrl_record_review_completed` were offered on a particular turn, it must inspect that actual model request's dispatched function list after filtering. Inventory presence alone proves neither dispatch nor a successful review-tool call.
144
144
>
145
-
> **Note on builtin tool filtering:**The `source: "builtin"`entries in `tools[]`reflect the _instance-level superset_ registered in `ToolRegistry`. Per-model filters (e.g. `apply_patch` only on gpt-\*, `codesearch`/`websearch` only for aictrl provider) are applied at dispatch time inside `resolveTools` and are **not** reflected here. Consumers should treat the presence of a builtin tool name as "registered and potentially available", not as "guaranteed to appear in the model's function list". The strong structural guarantee (tool present ↔ tool in dispatch list) applies only to `source: "mcp"` entries.
145
+
> **Inventory versus dispatch:**Builtin entries reflect the _instance-level superset_ registered in `ToolRegistry`; model-specific builtin filters are applied when resolving each request. MCP entries come from connected servers' resolved tool catalogs. Both sources can be excluded by effective work-tool selection, permissions or request-specific disabling. The shared MCP naming helper guarantees identifier consistency, not inclusion in the dispatched function list.
Copy file name to clipboardExpand all lines: README.md
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -63,7 +63,7 @@ Stdin is read only when a standalone `-` is a message operand, or when no messag
63
63
64
64
The marker can appear before or after `--`; repeated markers still read stdin once. Markers are removed from the model message, command arguments and derived session title. `--` ends option parsing, so `run -- -` also selects stdin. An operand such as `"literal -"` remains ordinary text. An explicit empty operand (`run ""`) does not select implicit stdin. Selected stdin is read to EOF and appended after one actual newline. Ordinary prompt operands are joined with spaces without adding artificial quotes or backslash escapes; quotes, backslashes and whitespace inside an operand remain its text.
65
65
66
-
Option values belong to their options: `--model=-`or`--command=-`does not select stdin, and scalar options also accept a separated hyphen value. Use `--file=-` to attach a file literally named `-`; `--file -` is missing its required file value. Multiple file values and repeated `--skill` flags retain their existing behavior. Required option values are checked by the argument parser; explicit `--title ""` remains supported.
66
+
Scalar required-value options own a separated hyphen: `--model-`and`--command-`do not select stdin, nor do their `=` forms. Optional `--port` leaves a separated `-` as the stdin operand (`--port -`). Use `--file=-` to attach a file literally named `-`; `--file -` is missing its required file value. Multiple file values and repeated `--skill` flags retain their existing behavior. Required option values are checked by the argument parser; explicit `--title ""` remains supported.
67
67
68
68
An explicit marker reads stdin even from a terminal. Any selected stream can wait for EOF indefinitely; this change adds no stdin timeout. Callers must close selected input or apply their own supervisor deadline.
Copy file name to clipboardExpand all lines: RELEASING.md
+32-9Lines changed: 32 additions & 9 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -10,6 +10,29 @@
10
10
bun test test/cli/run-mcp-discovery.test.ts test/mcp/discovery-recovery.test.ts test/cli/publish-workflow.test.ts test/cli/run-provider-finish.test.ts test/cli/run-signal-cancellation.test.ts test/cli/classify-session-error.test.ts test/session/idle.test.ts test/session/processor-idle.test.ts
11
11
```
12
12
13
+
Before publication, run the separate current-versus-previous native release
14
+
comparison. Prepare and review explicit candidate and previous published binary
15
+
artifacts, their SHA256/version/provenance pins and one existing immutable
16
+
fixture image using [the manifest contract](script/token-budget/release-fixtures/README.md).
17
+
Provision the pinned Python counter in a fresh directory (requires Python3.12
18
+
and network access during setup only), then capture and count offline:
0 commit comments