Repository navigation
Develop - #12
Merged
Merged
Develop#12
Conversation
sync: merge main into claude-developer-1
Implements entropy-based automatic redaction as a complement to pattern-based redaction. High-entropy tokens (API keys, tokens, random secrets) are detected statistically and replaced with [REDACTED] before sending commands to LLM providers. - internal/redact/entropy.go: Entropy() function + EntropyRedactor struct - internal/redact/entropy_test.go: tests for entropy calc and redaction behavior - internal/config/config.go: per-provider entropy_redaction_enabled field, global entropy_threshold/entropy_min_length config, UpdateProviderEntropyRedaction() - internal/llm/multi.go: entropy redactor applied per-provider in Query() - redact_cmd.go: guard-sh redact entropy on/off subcommand - main.go: entropy redactor init, entropy state in status output - help.go: entropy redact commands documented - config.default.yaml: entropy_threshold (4.5) and entropy_min_length (20) defaults Closes #10 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Restructures entropy-based redaction config to match the per-provider nested format specified in issue #10: providers: claude: redact_entropy: enabled: true threshold: 4.5 min_length: 20 - EntropyConfig struct nested under ProviderConfig.RedactEntropy - Removed flat global entropy_threshold/entropy_min_length from Config - Per-provider EntropyRedactor instances (each with its own threshold/min_length) - UpdateProviderEntropyRedaction creates/updates nested redact_entropy block - Multi uses map[string]*EntropyRedactor instead of shared redactor + enabled map - Status shows threshold and min_length when entropy is on Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adopts the config format requested in PR #11 review comment: redaction: # global (mandatory) pattern_based: enabled: true patterns: [...] shannon_entropy_based: enabled: false threshold: 4.5 min_length: 20 providers: claude: redaction: # optional per-provider override shannon_entropy_based: enabled: true threshold: 4.0 min_length: 16 Changes: - New PatternRedactionConfig, EntropyRedactionConfig, RedactionConfig structs - Config.Redaction replaces flat RedactPatterns field - ProviderConfig.Redaction replaces PatternBasedRedactionEnabled + RedactEntropy - EffectivePatternRedaction/EffectiveEntropyRedaction merge global + per-provider - Multi uses patternRedactors map (per-provider) instead of shared redactor - UpdateProviderPatternRedaction/UpdateProviderEntropyRedaction write nested YAML - README.org redaction section fully rewritten - config.default.yaml updated to new format Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
install.sh was checking for the exact source path, so running it from a different directory would always add a new entry. Now it detects any existing guard-sh integration in the rc file and skips if found. Also copies the shell script to the config dir so the sourced path is stable. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Debug output now shows pattern/entropy on|off for every provider, even when no redaction occurs. Redacted command is shown beneath the provider line when the command was modified. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Each provider now shows the command state after pattern redaction and after entropy redaction separately, with "unchanged" when the stage produced no modifications. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
feat: Shannon entropy-based redaction
install.sh was unconditionally overwriting ~/.config/guard-sh/prompt.txt on every run. Now it skips the copy if the file already exists, matching the existing behavior for config.yaml. Closes #14 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…erwrite fix: preserve custom prompt.txt on reinstall
Cache is moved from guard.Guard into llm.Multi. Each provider now has its own cache entry keyed as "providerName:command". This ensures: - Provider 1 failing and Provider 2 answering does not cause Provider 1 to be skipped on the next run. - Reordering or removing a provider does not serve stale responses. Closes #13 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When send_working_directory is true (default), guard-sh now sends: Working directory: /etc/nginx Command: chmod 777 nginx.conf to the LLM instead of the bare command. The cache key includes the working directory so identical commands in different directories are evaluated independently. Set send_working_directory: false in config to revert to command-only behaviour. Closes #16 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two edge cases were untested and broken: 1. Multiline commands — the widget was firing on the first line before the full command was assembled. Now checks $CONTEXT: if it is 'cont' (zsh waiting for more input due to open quotes, backslash, braces, if/for/while bodies etc.), the Enter is passed through unchanged. guard-sh intercepts only when $CONTEXT is 'start' (complete input). 2. History expansion — !! / !$ / !cmd tokens were passed raw to guard-sh, so it evaluated a different command than what would actually run. Now calls `zle .expand-history` before reading $BUFFER, so the expanded form is checked. Closes #15 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…cases fix: handle multiline commands and history expansion in zsh integration
…aware-cache feat: per-provider cache keys to prevent cross-provider cache hits
- Add shell/guard.fish with full fish integration (bind \n/\r, commandline widget, guard-sh on/off/--global) - Embed guard.fish in binary via //go:embed - Update setup.go to write guard.fish and handle fish RC file ($XDG_CONFIG_HOME/fish/config.fish) - Update install.sh with fish case - Add TestSetup_FreshInstall_Fish; update UnsupportedShell test to use tcsh Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- status command now shows work dir on/off - README.org: add send_working_directory to config example and add dedicated Working directory context section with usage and example - CLAUDE.md: update request flow, shell integration, config sections - guard_test.go: add TestCheck_QuerySentToLLM and TestCheck_WhitelistUsesRawCmd Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- healthcheck now checks fish RC file (~/.config/fish/config.fish) - README.org: add fish to requirements, install descriptions, shell support table, healthcheck example, and setup description - CLAUDE.md: update project description, request flow, and shell integration section to include fish Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…gration feat: add fish shell integration
- Remove cache from guard.go (moved to llm.Multi in #13) - Update guard_test.go: use new New() signature (no cacheMaxSize), keep rawCmd/query Check() signature, drop cache tests moved to multi Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…irectory feat: include working directory in LLM evaluation and cache key
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
1 participant
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.
No description provided.