Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 7 additions & 0 deletions .changes/bugs/fix-6916-upstream-pointers-a3b43c16.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
---
bump: patch
---

### Fixed

- `/bugs:scan` gives its `opus` fallback a reason, says its run metadata names the model asked for rather than proving the model that ran, and points at the sub-agents model section with an as-of date and recheck trigger.
7 changes: 7 additions & 0 deletions .changes/context-guard/fix-6916-upstream-pointers-96bf3770.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
---
bump: patch
---

### Fixed

- The reader contract's subagent `autoCompactWindow` row points at the sub-agents docs section that documents the field, in place of the changelog, and no longer says the surface appears only there.
7 changes: 7 additions & 0 deletions .changes/guardrails/fix-6916-upstream-pointers-c287750c.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
---
bump: patch
---

### Fixed

- The `workflow-resilience-check` description says the fan-out it flags has no wave cap of its own, in place of calling it un-throttled, and points at the workflow runtime's concurrency limit.
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
---
bump: patch
---

### Fixed

- The `evaluator` agent's model and effort pins carry their reasons, and the effort pin points at the upstream effort table with an as-of date and recheck trigger.
15 changes: 12 additions & 3 deletions .github/actions/resolve-config/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -121,8 +121,13 @@ the `limits` ([README, Limits](../../../docs/conventions/pr-pipeline/README.md#l
that profile's `model`, `effort` (null when unset), `subagent-model` (`inherit` when unset) and
`max-turns`, from its activity's route, else `models.default-profile`, else, with no `models`
block, the built-in `standard` profile (model `opus`, 75 turns); a model outside `haiku`,
`sonnet` and `opus` resolves to `opus`; at input `model-rung` 1 the
profile's `escalate-to` replaces it, one rung at most; a script slot carries `models`
`sonnet` and `opus` resolves to `opus`. That excludes the frontier aliases `fable` and `best` on
purpose: the lanes run no frontier model until the owner sets a Fable allowance for them, and the
schema refuses both on a profile (`pr-pipeline.schema.json`, `model` and `subagent-model`), so
this rule catches a config that reached the resolver unchecked. The upstream alias list is
[model-config, Model aliases](https://code.claude.com/docs/en/model-config#model-aliases), as of
2026-10-31; recheck when an alias is added or retired. At input `model-rung` 1 the
profile's `escalate-to` replaces that profile, one rung at most; a script slot carries `models`
null. A skill slot with input `issue`, in a config with `models.issue-dimensions`, also carries
`issue-dimensions` and `timeout-minutes`, and its `max-turns` is the dimensions' turn cap; its
Complexity's `complexity-profiles` entry, when there is one, replaces the route's profile. The
Expand Down Expand Up @@ -171,7 +176,11 @@ issue dimensions. `args` is free text and stays in the file.

`claude-args` is `--model <alias>`, then `--max-turns <n>`, then
`--effort <level>` when the profile sets one; `settings-json` is
`{"env":{"CLAUDE_CODE_SUBAGENT_MODEL":"<alias>"}}`, or empty for `inherit`. Both are built only
`{"env":{"CLAUDE_CODE_SUBAGENT_MODEL":"<alias>"}}`, or empty for `inherit`, which upstream treats
the same as unset. The variable is a default a spawn's `model` or an agent's frontmatter outranks:
[sub-agents, Choose a model](https://code.claude.com/docs/en/sub-agents#choose-a-model), as of
2026-10-31; recheck when that section changes the variable's rank or its `inherit` value.
`claude-args` and `settings-json` are built only
from schema enums and an integer, and the step fails on any other `claude-args` shape. `model` and
`effort` are the same values for a caller on ci-workflows' `run-claude-cli`: `model` is the alias;
`effort` is empty when the profile sets none. A script gets all six model outputs empty.
Expand Down
11 changes: 8 additions & 3 deletions plugins/bugs/skills/scan/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -158,14 +158,19 @@ resolved effort is lower, pass no `effort`.

When `/multi-agent:route` does not resolve, omit the model for both stages so they inherit the
session's, except when the session model is frontier or unknown, where pass `opus`; say once in the
report that enabling the multi-agent plugin makes this routing configurable. With no role map, gates
pass no `effort` either. Pointer: for the per-spawn
report that enabling the multi-agent plugin makes this routing configurable. The `opus` fallback is
this skill's stand-in for the fan-out guard `/multi-agent:route` applies: it keeps a wide fan-out
off the frontier tier; which model the alias then runs on is the harness's call (model pointer
below). With no role map, gates pass no `effort` either. Pointer: for the per-spawn
`effort` and its precedence, see <https://code.claude.com/docs/en/sub-agents#choose-an-effort-level>.
As of: 2026-10-10. Recheck trigger: that section changes precedence or which spawns honor `effort`.

Name the model and effort each stage was dispatched with in the run metadata: the level passed, or
`session` when none was. A dispatch the harness rejects means the role map names an alias this
session cannot use: report it rather than substitute one.
session cannot use: report it rather than substitute one. The metadata names the model asked for,
not proof of the model that ran. Pointer: for how a spawn's `model` alias resolves and what happens to a blocked
one, see <https://code.claude.com/docs/en/sub-agents#choose-a-model>. As of: 2026-10-31. Recheck
trigger: that section changes how an alias resolves or how a blocked value is handled.

**Breadth by scope.** After Step 1 enumerates the files (test suites excluded), classify the scope:

Expand Down
5 changes: 3 additions & 2 deletions plugins/context-guard/reference/reader-contract.md
Original file line number Diff line number Diff line change
Expand Up @@ -466,7 +466,7 @@ precedence.
| Surface | Kind | What this plugin relies on | Pointer |
|---|---|---|---|
| `autoCompactWindow` | settings key (top level or per model) | A token count that moves the trigger for the model it applies to. Unset, the default is the docs' default-thresholds table (pointer above); set where we cannot see it, we assume no number (below). Normalize it into the percentage shape before comparing (below). | [settings-reference: `autoCompactWindow`](https://code.claude.com/docs/en/settings-reference#autocompactwindow); [settings-reference: `modelSettings`](https://code.claude.com/docs/en/settings-reference#modelsettings); [model-config: Set the auto-compact window](https://code.claude.com/docs/en/model-config#set-the-auto-compact-window) |
| `autoCompactWindow` | subagent frontmatter | A subagent can carry its own window (Claude Code 2.1.296). The snapshot cannot see that window, and the surface is in the changelog only and not probed, so we never assume a subagent shares the session's trigger. | [Claude Code changelog](https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md) |
| `autoCompactWindow` | subagent frontmatter | A subagent can carry its own window. The snapshot cannot see that window, and the surface is not probed, so we never assume a subagent shares the session's trigger. | [sub-agents: Set the auto-compact window for one subagent](https://code.claude.com/docs/en/sub-agents#set-the-auto-compact-window-for-one-subagent) |
| `CLAUDE_CODE_AUTO_COMPACT_WINDOW` | environment variable | Read as the effective window whenever it is set, ahead of the setting, the command, and the flag. | [env-vars: Variables](https://code.claude.com/docs/en/env-vars#variables) |
| `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE` | environment variable | Read as able only to move the trigger earlier, never later. | [env-vars: Variables](https://code.claude.com/docs/en/env-vars#variables) |
| `autoCompactEnabled` / `DISABLE_AUTO_COMPACT` | `settings.json` key / environment variable | Either one turning auto-compact off leaves the dumb band as the only tripwire. `DISABLE_AUTO_COMPACT` leaves the manual `/compact` command available. | [settings-reference: `autoCompactEnabled`](https://code.claude.com/docs/en/settings-reference#autocompactenabled); [env-vars: Variables](https://code.claude.com/docs/en/env-vars#variables) |
Expand All @@ -476,7 +476,8 @@ We read a configured window above the model's context window as the model's wind
nothing.

- **Pointer**: per row above.
- **As of**: 2026-10-10 for the two `autoCompactWindow` rows; 2026-10-15 for the
- **As of**: 2026-10-10 for the settings `autoCompactWindow` row; 2026-10-31 for the subagent
frontmatter `autoCompactWindow` row; 2026-10-15 for the
`DISABLE_COMPACT` entry and the `DISABLE_AUTO_COMPACT` manual-`/compact` sentence; 2026-08-19
for the other rows.
- **Recheck trigger**: a release note or one of those sections changes a surface's units, range,
Expand Down
2 changes: 1 addition & 1 deletion plugins/guardrails/.claude-plugin/plugin.json
Original file line number Diff line number Diff line change
Expand Up @@ -159,7 +159,7 @@
"workflow_resilience_check_enabled": {
"type": "boolean",
"title": "Workflow-resilience-check guard",
"description": "Advises on un-throttled Workflow fan-out; never blocks. Off by default since 0.20.0: a behavioral-class prose injector, config-disabled per the instruction-economy evidence gate (#2021). Set true to opt back in.",
"description": "Advises on Workflow fan-out with no wave cap of its own; never blocks. Off by default since 0.20.0: a behavioral-class prose injector, config-disabled per the instruction-economy evidence gate (#2021). Set true to opt back in.",
"default": false
},
"flag_commit_pr_skill_bypass_enabled": {
Expand Down
Loading
Loading