@cohortapp/agent-sdk 2.18.27 → 2.18.28

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/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@cohortapp/agent-sdk",
3
- "version": "2.18.27",
3
+ "version": "2.18.28",
4
4
  "description": "Cohort Agent SDK — autonomous AI colleague runtime. Deploy senior AI colleagues on dedicated Mac minis, wired to the Cohort operating surface.",
5
5
  "type": "module",
6
6
  "bin": {
@@ -371,6 +371,31 @@ mkdir -p "$CLAUDE_SETTINGS_DIR"
371
371
  # RCE vector — and have been removed entirely. Collective-memory hooks are
372
372
  # wired separately and idempotently by `maestro global-setup` below, which
373
373
  # deep-merges its own hooks.)
374
+ # NO OUTPUT-STYLE PLUGINS ON A SEAT.
375
+ #
376
+ # explanatory-output-style and learning-output-style were both enabled in the
377
+ # enabledPlugins block below, and both are wrong for an agent seat: a seat does
378
+ # not write for a developer at a terminal, it posts to colleagues in a room.
379
+ #
380
+ # explanatory instructs the model to wrap its work in decorative teaching
381
+ # blocks. Observed 2026-09-30 20:12:55Z: a seat posted one verbatim into a
382
+ # shared engineering channel, backticks and box-drawing rules included, in the
383
+ # middle of a status message.
384
+ #
385
+ # learning instructs the model to stop and ask the reader to write 5-10 lines
386
+ # of code, marked with a TODO. A seat answering a colleague must never hand
387
+ # back homework.
388
+ #
389
+ # Enabling the plugin is what turns the style ON, not merely what makes it
390
+ # available: the machines exhibiting this carry no separate outputStyle key, and
391
+ # a session on a box with both enabled reports itself in learning mode. So
392
+ # listing them here was a fleet-wide activation.
393
+ #
394
+ # This fixes seats provisioned from here on. An existing seat still carries the
395
+ # flags in its own global settings file and has to be cleaned there. The upgrade
396
+ # path deliberately does not rewrite a machine global settings file, because
397
+ # silently editing 16 operator configs is not a repair an upgrade is entitled to
398
+ # make; the remediation is a one-line edit the operator runs.
374
399
  MAESTRO_SETTINGS_ADDITIONS='{
375
400
  "$schema": "https://json.schemastore.org/claude-code-settings.json",
376
401
  "enabledPlugins": {
@@ -398,8 +423,6 @@ MAESTRO_SETTINGS_ADDITIONS='{
398
423
  "remember@claude-plugins-official": true,
399
424
  "mcp-server-dev@claude-plugins-official": true,
400
425
  "zapier@claude-plugins-official": true,
401
- "explanatory-output-style@claude-plugins-official": true,
402
- "learning-output-style@claude-plugins-official": true,
403
426
  "claude-mem": true,
404
427
  "claude-pace": true
405
428
  },