Fix: Minor logging uplift for debugging of prompt injection mitigation#7195
Merged
dorien-koelemeijer merged 3 commits intomainfrom Feb 17, 2026
Merged
Fix: Minor logging uplift for debugging of prompt injection mitigation#7195dorien-koelemeijer merged 3 commits intomainfrom
dorien-koelemeijer merged 3 commits intomainfrom
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
This PR updates the prompt-injection scanning internals to surface (via logging) whether pattern-based scanning was used as a fallback when ML-based command injection detection isn’t available or fails.
Changes:
- Add a
used_pattern_detectionflag toDetailedScanResultto track when pattern-based scanning was used. - Switch the
tracing::info!fieldhas_patternsto report the fallback-path usage rather than presence of pattern matches. - Propagate
used_pattern_detectionthrough intermediate scan results used to build the final explanation/logging.
Comment on lines
243
to
247
| confidence: max_confidence, | ||
| pattern_matches: Vec::new(), | ||
| ml_confidence: Some(max_confidence), | ||
| used_pattern_detection: false, | ||
| }) |
There was a problem hiding this comment.
scan_conversation always sets ml_confidence: Some(max_confidence) even if every classifier call failed (all scan_with_classifier results were None), which makes downstream logic treat this as a real ML signal (and currently reduces tool_confidence by 10% when the value is 0.0). Track whether any classification succeeded (e.g., fold an Option<f32> or keep a success flag) and return ml_confidence: None when there were no successful results.
Comment on lines
+80
to
+81
| monotonic_counter.goose.security_command_classifier_enabled = if command_classifier_enabled { 1 } else { 0 }, | ||
| monotonic_counter.goose.security_prompt_classifier_enabled = if prompt_classifier_enabled { 1 } else { 0 }, |
There was a problem hiding this comment.
Using monotonic_counter.* = if enabled { 1 } else { 0 } is likely to produce confusing metrics (a counter with value 0 is typically a no-op and the name suggests a gauge); consider logging the booleans as normal fields (e.g. command_classifier_enabled = ...) and emitting a separate monotonic_counter metric with value 1 (or separate enabled/disabled counters) so disabled configurations are still observable.
Suggested change
| monotonic_counter.goose.security_command_classifier_enabled = if command_classifier_enabled { 1 } else { 0 }, | |
| monotonic_counter.goose.security_prompt_classifier_enabled = if prompt_classifier_enabled { 1 } else { 0 }, | |
| monotonic_counter.goose.security_classifier_configuration_logged = 1, | |
| security_command_classifier_enabled = command_classifier_enabled, | |
| security_prompt_classifier_enabled = prompt_classifier_enabled, |
michaelneale
approved these changes
Feb 17, 2026
jh-block
added a commit
that referenced
this pull request
Feb 18, 2026
* origin/main: (49 commits) chore: show important keys for provider configuration (#7265) fix: subrecipe relative path with summon (#7295) fix extension selector not displaying the correct enabled extensions (#7290) Use the working dir from the session (#7285) Fix: Minor logging uplift for debugging of prompt injection mitigation (#7195) feat(otel): make otel logging level configurable (#7271) docs: add documentation for Top Of Mind extension (#7283) Document gemini 3 thinking levels (#7282) docs: stream subagent tool calls (#7280) Docs: delete custom provider in desktop (#7279) Everything is streaming (#7247) openai: responses models and hardens event streaming handling (#6831) docs: disable ai session naming (#7194) Added cmd to validate bundled extensions json (#7217) working_dir usage more clear in add_extension (#6958) Use Canonical Models to set context window sizes (#6723) Set up direnv and update flake inputs (#6526) fix: restore subagent tool call notifications after summon refactor (#7243) fix(ui): preserve server config values on partial provider config save (#7248) fix(claude-code): allow goose to run inside a Claude Code session (#7232) ...
aharvard
added a commit
that referenced
this pull request
Feb 18, 2026
* origin/main: feat: add GOOSE_SUBAGENT_MODEL and GOOSE_SUBAGENT_PROVIDER config options (#7277) fix(openai): support "reasoning" field alias in streaming deltas (#7294) fix(ui): revert app-driven iframe width and send containerDimensions per ext-apps spec (#7300) New OpenAI event (#7301) ci: add fork guards to scheduled workflows (#7292) fix: allow ollama input limit override (#7281) chore: show important keys for provider configuration (#7265) fix: subrecipe relative path with summon (#7295) fix extension selector not displaying the correct enabled extensions (#7290) Use the working dir from the session (#7285) Fix: Minor logging uplift for debugging of prompt injection mitigation (#7195) feat(otel): make otel logging level configurable (#7271) docs: add documentation for Top Of Mind extension (#7283) Document gemini 3 thinking levels (#7282) docs: stream subagent tool calls (#7280) Docs: delete custom provider in desktop (#7279) # Conflicts: # ui/desktop/src/components/McpApps/McpAppRenderer.tsx
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
2 participants
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.
Summary
Type of Change
AI Assistance
Testing
Local/manual testing.