Skip to content
Open
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
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
---
bump: patch
---

### Changed

- `draft-auto-mode-rules` now offers the `allow` rules that `/auto-mode-setup` (Claude Code 2.1.329 and later) drafts from recent auto mode denials as input to the `allow` section. Its reference records what the command reads and writes.
14 changes: 10 additions & 4 deletions plugins/harness-config/skills/draft-auto-mode-rules/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -56,8 +56,12 @@ can mean either.
If /auto-mode-setup is available in your session (gate basis: its row of the commands reference;
records in [reference/native-auto-mode-setup.md](reference/native-auto-mode-setup.md)), you can
run it instead of or alongside this skill to have the `environment` section drafted from your sessions
and saved for you; this interview then covers the allow and deny sections. If they take it, run
Phase 1 again after it saves so the draft builds on what it wrote. When the caller passes
and saved for you; this interview then covers the allow and deny sections. It can also offer `allow`
rules drawn from your recent auto mode denials (version in the records). Those rules are input to
the `allow` section here: paste the ones you accepted, or the `allow` list `claude auto-mode config`
prints, and the interview asks for each one's exclusions and transcript-visible condition instead of
drafting it from recall. If they take it, run Phase 1 again after it saves so the draft builds on
what it wrote. When the caller passes
`--unattended`, the run records the offer in its output instead of asking;
without that flag it asks.

Expand All @@ -78,8 +82,10 @@ section to focus the interview on. `--unattended`, anywhere in the arguments, is
declaring that nobody is present, which governs the `/auto-mode-setup` offer under "Routing". Take
it out first, then read the rest.

There is no history input: a read of "the repo's observed prompt and denial history" names no actual
location, and an unnamed read surface in a skill shipped to consumers is unreviewable.
This skill reads no history itself: a read of "the repo's observed prompt and denial history" names
no actual location, and an unnamed read surface in a skill shipped to consumers is unreviewable.
Denial history reaches the draft only through the `allow` rules `/auto-mode-setup` saved, which the
person brings to the interview (see "Routing").

## Phase 1: Read what already exists

Expand Down
Original file line number Diff line number Diff line change
@@ -1,21 +1,26 @@
# The built-in `/auto-mode-setup` command, as this skill relates to it

Four-part records behind the `## Boundary` section in `SKILL.md`. The section carries the
conclusion; this file carries what it rests on. Nothing here asserts that the command is present in
any session.
Records behind the `## Boundary` section in `SKILL.md`. The section carries the conclusion; this
file carries what it rests on. A row with an As of date stores a copied string; the other rows are
pointers and carry no stamp. Nothing here asserts that the command is present in any session.

| Claim | Basis | As of | Recheck when |
|---|---|---|---|
| `/auto-mode-setup` is a built-in command, user-invocable only, hidden from the command menu, and gated; its description is "Teach auto mode about your environment, plus optional rule tweaks" | The `/harness-ops:inventory` extraction of the installed 2.1.284 binary (`model_invocable: false`, `user_invocable: true`, `hidden: true`, `gated: true`) | 2026-09-29, Claude Code 2.1.284 | A release renames or removes the command, or changes its invocability or hidden flag |
| We treat it as the writer of `autoMode.environment` entries, saved after the person reviews them, and this skill as the reader; what it drafts and where it saves are read from its row | Pointer: when routing, fetch the `/auto-mode-setup` row of [commands: all commands](https://code.claude.com/docs/en/commands#all-commands) live | 2026-10-11 | The commands page row changes what it drafts or where it saves |
| Its availability depends on plan and version requirements, which we never restate: read them from the same row before telling a person it is available | Same pointer | 2026-10-11 | The commands page row changes its plan or version requirements |
| We treat it as the writer of auto mode classifier entries, saved after the person reviews them, and this skill as the reader | Pointer: when routing, fetch the `/auto-mode-setup` row of [commands: all commands](https://code.claude.com/docs/en/commands#all-commands) live; when deciding what it drafts, fetch [auto mode config: generate environment entries](https://code.claude.com/docs/en/auto-mode-config#generate-environment-entries) live | | |
| Its availability depends on plan and version requirements, which we never restate: read them from the same row before telling a person it is available | Pointer: when telling a person it is available, fetch the same row live | | |
| We treat what it reads as possible input to the interview, never as a source this skill reads itself | Pointer: when deciding what it read, fetch [auto mode config: what `/auto-mode-setup` reads](https://code.claude.com/docs/en/auto-mode-config#what-auto-mode-setup-reads) live | | |
| Since Claude Code 2.1.329 it can offer `allow` rules drawn from recent auto mode denials; we treat those rules as input to the interview's `allow` section. No auto mode config section covers the denial scan yet | Changelog, Claude Code 2.1.329: "Fixed `/auto-mode-setup` ignoring recent auto mode denials when it scans a project for rules to suggest; it may now offer allow rules from them, and you review each before it's saved" ([changelog](https://code.claude.com/docs/en/changelog)) | 2026-11-21, Claude Code 2.1.329 | The auto mode config page documents the denial scan, or a release changes what it offers |
| The changelog and the auto mode config page disagree on how the draft is reviewed | [changelog](https://code.claude.com/docs/en/changelog) and [auto mode config: generate environment entries](https://code.claude.com/docs/en/auto-mode-config#generate-environment-entries) | | |
| We treat its save as the user-settings write that Phase 1 then reads | Pointer: when deciding what it writes, where it saves, or how it handles `"$defaults"` and `permissions.allow`, fetch [auto mode config: generate environment entries](https://code.claude.com/docs/en/auto-mode-config#generate-environment-entries) live, including its "Review and save the draft" subsection | | |
| Its argument hint is `[--request-id <uuid>] (--wizard posture=... scope=... depth=... --propose \| --expect-sha256 <64-hex> --apply-file <path>)`: a propose step and a hash-checked apply step | The `/harness-ops:inventory` extraction of the installed 2.1.284 binary | 2026-09-29, Claude Code 2.1.284 | A release changes the argument hint |

## Why the verdict is complementary

The command writes: it produces `environment` entries and saves them to user settings after the
person reviews them. The commands page documents it for `environment` entries only; its "optional
rule tweaks" are not described there. This skill never writes and covers every classifier section,
with an interview for stated exclusions and a condition the classifier can see in the transcript.
Using the command for `environment` and this skill for the rest composes cleanly, since the
command's save is what Phase 1 then reads.
We treat the command as the writer of classifier entries to user settings, once the person reviews
its draft. Since 2.1.329 its `allow` entries can come from recent denials (the stored changelog
line above), which is the allow-rule evidence this skill's interview has no source for. This
skill never writes and covers every classifier section, with an interview for stated exclusions and
a condition the classifier can see in the transcript. Using the command for `environment` and
denial-based `allow` rules, then this skill to sharpen those rules and draft the rest, composes
cleanly, since the command's save is what Phase 1 then reads.
Loading