Conversation
The shared .claude/settings.json pinned an AWS profile, forced Bedrock, and hard-coded five model IDs. Those values are specific to one machine, so every other contributor inherits a profile that does not exist for them and a model set that goes stale on each release. It also set MAX_THINKING_TOKENS to 1024 and CLAUDE_CODE_MAX_OUTPUT_TOKENS to 10240, which throttles reasoning and output for everyone who clones the repo. Move those values to a new .claude/settings.local.json.example template. Contributors copy it to .claude/settings.local.json, which is already gitignored and takes precedence over the shared file. Keep the telemetry opt-out, which is machine-neutral. Also nest the permission rules under a "permissions" object. The rules were top-level "allow" and "deny" keys, which is not the schema Claude Code documents, so the deny list was not taking effect. All 81 deny and 69 allow entries are unchanged, as are the hooks and the status line. Skills, hooks, rules, scripts, and the status line are separate tracked files and continue to ship to everyone who clones the repo. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Remove the default AWS_PROFILE reference from the check-claude-settings workflow and link to settings.local.json.example instead of an inline code block. Update the settings comment, example file, and AI tools page to match. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Describe what the shared Claude settings contain, note that the example local settings target Bedrock, and add a section on truncated output. Add a GitHub Copilot section. Update the example file to current model IDs and drop the token limit keys. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Make the GitHub Copilot section an H3 under Using AI Assistants, note that Copilot users use prompt files instead of skills, clarify the provider setup sentence, and name model selection in the workflow comment. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This branch has not been deployed
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Context
This addresses the
#docsthread about whether.claude/belongs in.gitignore.Both positions in that thread are right, because they are about different files. Skills are not in
settings.json. They are seven standalone tracked files, and they keep shipping to everyone who clones the repo no matter what happens here:So gitignoring
.claude/is not the answer — that would unshare the skills, which is exactly what we want to keep. Only one part ofsettings.jsoncauses the problem, and this PR moves just that part.What breaks today
The
envblock pinned values that are specific to one machine:AWS_PROFILE: my-sandboxCLAUDE_CODE_USE_BEDROCK: 1ANTHROPIC_MODEL+ 4 more model IDsclaude-sonnet-4-6/claude-opus-4-8; goes stale every releaseMAX_THINKING_TOKENS: 1024CLAUDE_CODE_MAX_OUTPUT_TOKENS: 10240OTEL_METRICS_EXPORTER: otlpThe last two are the least visible and probably the most damaging: they quietly degrade the agent for everyone.
What changed
.claude/settings.local.json.example. Copy it to.claude/settings.local.json, which is already gitignored and takes precedence over the shared file. Nothing is lost — the Bedrock setup is still onecpaway.permissions. They were top-levelallowanddenykeys. The documented schema is"permissions": { "allow": [...], "deny": [...] }and no top-level form exists, so the deny list was almost certainly inert — meaning thesudo,kubectl apply,terraform apply, andrm -rf /guards were not actually blocking anything, while.claude/rules/security.mdstates that guardrails are mandatory. Worth confirming with/permissionsafter merge.All 81 deny and 69 allow entries are byte-identical, verified programmatically; only the nesting changed. Hooks and status line are untouched. Both files parse as valid JSON.
Note for reviewers
check-claude-settings.ymlfails any PR touching.claude/**unless the author is on its tech-writer allowlist, so this PR will go red and post the standard warning comment. That gate is working as intended — this change does need TW sign-off, which is why it is a separate PR from the docs work rather than bundled in.The permissions nesting fix is arguably the more urgent half. Happy to split it into its own PR if you would rather land that first and discuss the
envquestion separately.🤖 Generated with Claude Code