foldrun 20.9.2 → 20.9.4

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/docs/providers.md CHANGED
@@ -195,6 +195,43 @@ those families on both wires and says so in the drop line. Every other id
195
195
  keeps it, because a gateway ignores a field it cannot use and an o-series,
196
196
  gpt-5, grok or deepseek-r model wants it.
197
197
 
198
+ ### Persian prompts, by tier (checked 11 Nov 2026)
199
+
200
+ An eval of eight Persian tasks, compared on the three tiers with **Compare
201
+ models** ([Evals](evals#comparing-models)). Each case pairs a deterministic
202
+ check, where one fits, with a judge on `max` (Opus, low effort). The subject
203
+ was a plain agent told to write natural Persian, keep code in Latin script
204
+ and use Persian digits:
205
+
206
+ | case | fast (Haiku) | default (Sonnet) | max (Opus) |
207
+ |---|---|---|---|
208
+ | half-spaces in a short paragraph | pass | pass | pass |
209
+ | a formal notice to a customer | fail | pass | pass |
210
+ | Persian digits and thousands separators | pass | pass | pass |
211
+ | a command kept intact in Persian text | pass | pass | pass |
212
+ | a one-sentence summary | pass | pass | pass |
213
+ | a technical translation from English | pass | pass | pass |
214
+ | exactly three numbered points | pass | fail | pass |
215
+ | Persian when asked in English | fail | pass | pass |
216
+ | **passed** | **6 / 8** | **7 / 8** | **8 / 8** |
217
+ | cost per run · time | $0.013 · 6.7 s | $0.015 · 6.5 s | $0.024 · 8.4 s |
218
+
219
+ Every tier wrote correct Persian: half-spaces where they belong, Persian
220
+ digits when asked, code and commands left intact inside right-to-left text.
221
+ The three failures were following the task, not the language:
222
+
223
+ - `fast` opened the formal notice with a line about itself and invented a
224
+ Persian calendar date the task never gave;
225
+ - `fast` made one grammatical slip («سامانه نام دامنه … رکوردهایی هستند»);
226
+ - `default` added an introductory line and ran one list item past its
227
+ ten-word limit.
228
+
229
+ So `default` is a sound default for Persian, and an agent whose output must
230
+ follow a strict shape (a fixed format, an exact length) earns `max`, or a
231
+ `verify:` that checks the shape. One run per case is a small sample — run
232
+ the comparison on your own agent's cases with `repeat: 3` before moving a
233
+ step to `fast`. The whole comparison cost $0.42 (runs and judges together).
234
+
198
235
  ## Why a turn stopped
199
236
 
200
237
  Every model reply ends with a stop reason, and the model loop acts on it: run
package/docs/secrets.md CHANGED
@@ -219,6 +219,10 @@ PATCH /api/secrets { "name": "DATAFORSEO_PASSWORD", "sites": ["api.dataforseo.
219
219
  PUT /api/secrets { "name": "…", "value": "…", "sites": ["api.dataforseo.com"] } # set and bind at once
220
220
  ```
221
221
 
222
+ A **remote MCP server**'s `headers:` and `url:` work the same way: bind the
223
+ key to the server's host and the sandbox's copy of them holds `${NAME}`,
224
+ filled by the proxy on the way to that host alone ([MCP](tools#mcp-someone-elses-tools)).
225
+
222
226
  What a program in the sandbox writes is unchanged; it sends the placeholder
223
227
  where the value would go:
224
228
 
package/docs/tools.md CHANGED
@@ -146,6 +146,15 @@ Both transports the SDK supports: a local process (`command:`, with `args:` and
146
146
  exposes reaches the agent that grants it. `mcpServers:` in an agent's own
147
147
  frontmatter is the inline spelling of the same thing.
148
148
 
149
+ A remote server is called with MCP's Streamable HTTP transport, the one the
150
+ current spec names; the older SSE transport, which it replaced, is not
151
+ offered. `${NAME}` works in `url:` and `headers:` alike. A secret **bound to
152
+ the server's host** ([Bound to sites](secrets#bound-to-sites)) never enters
153
+ the sandbox: the sandbox's copy of the server's URL and headers holds
154
+ `${NAME}`, and the egress proxy fills it in requests to that host and no
155
+ other. An unbound secret is filled before the step starts, so the sandbox
156
+ holds its value — bind a key you would not hand the agent.
157
+
149
158
  ## The library — define it once
150
159
 
151
160
  ```
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "foldrun",
3
- "version": "20.9.2",
3
+ "version": "20.9.4",
4
4
  "description": "Agents are just folders. Write them, check them, run them — on your machine.",
5
5
  "license": "FSL-1.1-ALv2",
6
6
  "type": "module",
@@ -11,7 +11,7 @@
11
11
  "node": ">=22"
12
12
  },
13
13
  "dependencies": {
14
- "@foldrun/core": "^18.3.2"
14
+ "@foldrun/core": "^18.4.0"
15
15
  },
16
16
  "files": [
17
17
  "bin",