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
docs(skills): weekly Claude Code skills refresh against Claude Code 2.1.302 (2026-10-18) - #111
Weekly refresh of the two Claude Code skills against Claude Code 2.1.302
(release, published
2026-10-16; npm latest and next, stable is 2.1.292). The previous refresh (#104) was
verified against 2.1.299; this one covers 2.1.300, 2.1.301 and 2.1.302.
Basis for the changes:
The upstream CHANGELOG.md sections for 2.1.300–2.1.302 (read in full).
The code.claude.com/docs pages for settings reference, settings, permissions, permission
modes, hooks, MCP, sub-agents, tools, env vars, CLI reference, workflows and Google Vertex AI,
fetched 2026-10-18.
Claims that only the changelog supports are labeled "changelog only" in the references. Where
the docs page lags a changelog entry (fork mode defaults, claude -p starting mode, /advisor
on Claude Platform on AWS), the reference says which one it is following.
Type of change
New or updated skill
Documentation improvement
Changes made
skills/client-config-claudecode (now v2.4)
references/permissions.md: the MCP allow-rule naming for servers whose name contains __ or
ends in _ is now documented upstream, so that part is no longer "changelog only" (deny and
ask rules still are); adds the auto mode classifier model (Sonnet 5 by default, fallbacks as
the permission-modes page lists them). Changelog only: CLAUDE_CODE_AUTO_MODE_MODEL,
server-side auto mode compacting an over-long conversation, and claude -p/Agent SDK sessions
starting in default on third-party providers, with telemetry off, or on a fresh install.
references/hooks.md: prompt/agent hook model default reworded to match the docs.
Changelog only: agent hooks honor model (2.1.300), at most 12 prompt/agent hooks of one event
run at once (2.1.302), updatedInput is validated (2.1.300), setMode ignored in a Cowork
cloud PermissionRequest allow (2.1.300), plugin WorktreeCreate/WorktreeRemove run with --worktree (2.1.300).
references/mcp.md: adds the MCP docs page's rules for root-level anyOf/oneOf/allOf
schemas, invalid input schemas, and tool names over 128 characters (the skill had none of
these). Changelog only: a tool whose schema the API still rejects is sent without parameters
(2.1.300).
references/settings-schema.md: advisorModel row gains CLAUDE_CODE_DISABLE_ADVISOR_TOOL
and the /advisor-on-Claude-Platform-on-AWS discrepancy; idleCompaction row gains the /config toggle (changelog only).
evals/evals.json: new mcp-server-name-with-double-underscore case (not yet run; README
results predate it).
CHANGELOG.md, README.md: v2.4 entry and version stamp.
skills/agent-creator-claudecode
autoCompactWindow is now in the sub-agents docs table. Dropped its "changelog only" label in SKILL.md, README.md, the validator docstring and references/frontmatter-reference.md,
and documented its range (100000–1000000), the CLAUDE_CODE_AUTO_COMPACT_WINDOW
precedence, and the 200k pitfall.
scripts/validate-agent.py: warns when autoCompactWindow is outside 100000–1000000.
references/frontmatter-reference.md, references/tools-reference.md, references/troubleshooting.md: changelog only (2.1.302): fork mode now defaults on in claude -p, the Agent SDK, the desktop app and VS Code (CLAUDE_CODE_FORK_SUBAGENT=0 turns it
off); the desktop app and VS Code always background the main conversation's subagents. The
sub-agents docs page still says fork mode is off for -p and the SDK.
README.md: version stamp.
docs/
docs/skills/client-config-claudecode.md, docs/skills/agent-creator-claudecode.md:
version line updated to 2.1.302.
docs/harness/claudecode/claudecode-vertexai.md: every claim re-checked against the live
Google Vertex AI docs page; only the verification stamp changed.
Checked and unchanged
All 245 rows of the settings reference index are already in the references and validator.
Every hook event on the hooks docs page is already covered.
The sub-agents frontmatter table and the --agents field list match the validator's field
list; the background-subagent built-in tool list matches.
Checklist
mkdocs build --strict passes with no warnings
All fenced code blocks have a language specifier
All links resolve correctly
No hardcoded secrets or tokens
New pages added to mkdocs.yml nav (none added)
New skills follow the SKILL.md / README.md / QUICKREF.md structure (none added)
Related docs updated in this same PR
Tested with
Claude Code 2.1.302 docs and changelog. Local checks: validate_artifact_sync.py (clean), python -m unittest discover -s scripts (34 pass), validate-agent.py on agent files with autoCompactWindow inside and outside the range, validate-settings.py on a sample file with idleCompaction and advisorModel, mkdocs build --strict, markdownlint-cli2 on the changed
files. The installed CLI is 2.1.300, so nothing here was exercised against a running 2.1.302.
Add the 2.1.300-2.1.302 changes to the permissions, hooks, MCP and
settings references, labeling claims only the changelog supports.
Document the MCP schema and name rules and the auto mode classifier
model from the upstream docs. Add the mcp-server-name-with-double-
underscore eval, and stamp the README and CHANGELOG (v2.4).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
autoCompactWindow is now in the sub-agents docs table, so drop its
"changelog only" label and document its range and precedence. The
validator warns on a value outside 100000-1000000. Note the 2.1.302
fork mode and foreground/background default changes as changelog only,
since the docs page still describes the earlier behavior.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
….302
Every claim on the Vertex AI page was re-checked against the live
Google Cloud Agent Platform docs page; only the verification stamp changed.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…oot-level schema combinators
On a deployment without the remote configuration, Claude Code skips a
tool with a root-level anyOf/oneOf/allOf instead of flattening it.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…o schemas without another dialect
Claude Code applies the meta-schema check only when $schema is absent or
draft 2020-12; other dialects still get the property-name check.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…e with nested Agent calls
The 2.1.302 changelog runs a subagent's own Agent calls in the foreground
by default; qualify the "every subagent in the background" statements.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… names containing __
The agent-creator references document only the __ case for tools:, so
don't extend the trailing-underscore rule to it.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…r special server names
Names containing __ or ending in _ need the wildcard form with the whole
name; the generic MCP permission pattern said the bare rule matches all.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…mode override
The changelog says claude mcp serve never offers forks, so the
CLAUDE_CODE_FORK_SUBAGENT row can't claim to enable them everywhere.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The new eval requires an imprecise explanation: under the documented mcp__<server>__<tool> grammar, mcp__reports__v2 is a rule for server reports and tool v2, not a bare rule for a server named reports; mcp__reports is the bare rule for server reports. Requiring the current wording can mark a technically correct answer wrong. Make the expected output and assertion distinguish the server/tool parse while still requiring mcp__reports__v2__* for the reports__v2 server.
… the server/tool parse
mcp__reports__v2 parses as server reports and tool v2 under the
mcp__<server>__<tool> grammar; the assertion no longer requires the
looser "rule for a server named reports" wording.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Copilot's summary-only finding on the mcp-server-name-with-double-underscore eval (round 3, no thread): valid. Fixed in b369c07: the expected output and assertion now accept the server/tool parse (mcp__reports__v2 names server reports, tool v2) instead of requiring the looser wording; mcp__reports__v2__* is still required. Codex's code and security reviews of fe61e9c found nothing.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
What does this PR do?
Weekly refresh of the two Claude Code skills against Claude Code 2.1.302
(release, published
2026-10-16; npm
latestandnext,stableis 2.1.292). The previous refresh (#104) wasverified against 2.1.299; this one covers 2.1.300, 2.1.301 and 2.1.302.
Basis for the changes:
CHANGELOG.mdsections for 2.1.300–2.1.302 (read in full).code.claude.com/docspages for settings reference, settings, permissions, permissionmodes, hooks, MCP, sub-agents, tools, env vars, CLI reference, workflows and Google Vertex AI,
fetched 2026-10-18.
the docs page lags a changelog entry (fork mode defaults,
claude -pstarting mode,/advisoron Claude Platform on AWS), the reference says which one it is following.
Type of change
Changes made
skills/client-config-claudecode(now v2.4)references/permissions.md: the MCP allow-rule naming for servers whose name contains__orends in
_is now documented upstream, so that part is no longer "changelog only" (deny andask rules still are); adds the auto mode classifier model (Sonnet 5 by default, fallbacks as
the permission-modes page lists them). Changelog only:
CLAUDE_CODE_AUTO_MODE_MODEL,server-side auto mode compacting an over-long conversation, and
claude -p/Agent SDK sessionsstarting in
defaulton third-party providers, with telemetry off, or on a fresh install.references/hooks.md:prompt/agenthookmodeldefault reworded to match the docs.Changelog only: agent hooks honor
model(2.1.300), at most 12 prompt/agent hooks of one eventrun at once (2.1.302),
updatedInputis validated (2.1.300),setModeignored in a Coworkcloud
PermissionRequestallow (2.1.300), pluginWorktreeCreate/WorktreeRemoverun with--worktree(2.1.300).references/mcp.md: adds the MCP docs page's rules for root-levelanyOf/oneOf/allOfschemas, invalid input schemas, and tool names over 128 characters (the skill had none of
these). Changelog only: a tool whose schema the API still rejects is sent without parameters
(2.1.300).
references/settings-schema.md:advisorModelrow gainsCLAUDE_CODE_DISABLE_ADVISOR_TOOLand the
/advisor-on-Claude-Platform-on-AWS discrepancy;idleCompactionrow gains the/configtoggle (changelog only).evals/evals.json: newmcp-server-name-with-double-underscorecase (not yet run; READMEresults predate it).
CHANGELOG.md,README.md: v2.4 entry and version stamp.skills/agent-creator-claudecodeautoCompactWindowis now in the sub-agents docs table. Dropped its "changelog only" label inSKILL.md,README.md, the validator docstring andreferences/frontmatter-reference.md,and documented its range (
100000–1000000), theCLAUDE_CODE_AUTO_COMPACT_WINDOWprecedence, and the
200kpitfall.scripts/validate-agent.py: warns whenautoCompactWindowis outside100000–1000000.references/frontmatter-reference.md,references/tools-reference.md,references/troubleshooting.md: changelog only (2.1.302): fork mode now defaults on inclaude -p, the Agent SDK, the desktop app and VS Code (CLAUDE_CODE_FORK_SUBAGENT=0turns itoff); the desktop app and VS Code always background the main conversation's subagents. The
sub-agents docs page still says fork mode is off for
-pand the SDK.README.md: version stamp.docs/docs/skills/client-config-claudecode.md,docs/skills/agent-creator-claudecode.md:version line updated to 2.1.302.
docs/harness/claudecode/claudecode-vertexai.md: every claim re-checked against the liveGoogle Vertex AI docs page; only the verification stamp changed.
Checked and unchanged
--agentsfield list match the validator's fieldlist; the background-subagent built-in tool list matches.
Checklist
mkdocs build --strictpasses with no warningsmkdocs.ymlnav (none added)SKILL.md / README.md / QUICKREF.mdstructure (none added)Tested with
Claude Code 2.1.302 docs and changelog. Local checks:
validate_artifact_sync.py(clean),python -m unittest discover -s scripts(34 pass),validate-agent.pyon agent files withautoCompactWindowinside and outside the range,validate-settings.pyon a sample file withidleCompactionandadvisorModel,mkdocs build --strict,markdownlint-cli2on the changedfiles. The installed CLI is 2.1.300, so nothing here was exercised against a running 2.1.302.
🤖 Generated with Claude Code