Prompt guardrail: stop false 'can't track commitments' - #14
Merged
Conversation
…lip) Logs showed her tell the user "No, I cannot create or track commitments or subcommitments" — while tracking commitments fine the turns before and after. Bundling "track this commitment" with "can you do subcommitments?" made the model over-generalize the true "no subcommitments" into a false "can't track commitments at all". Add a guardrail to the commitment section: she tracks FLAT commitments (with recurrence), has no nested subcommitment/subtask feature, and must track-as-one + say so rather than deny the whole capability — plus the general rule to say what she CAN do instead of refusing a capability she has a tool for. Assisted-by: Claude Code (Opus 4.8)
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
From the logs: she told the user "No, I cannot create or track commitments or subcommitments" — while tracking commitments fine the turns immediately before and after. Not a tool gap (she has
add/list/complete/drop_commitment, and it's her core duty) — a grounding slip: bundling "track this commitment" with "can you do subcommitments?" made the model over-generalize the true "no subcommitments" into a false "can't track commitments."Fix: a guardrail in the commitment section of the system prompt — she tracks FLAT commitments (with recurrence), has no nested subcommitment/subtask feature, and must track-as-one + say so rather than deny the capability. Plus the general rule: say what you CAN do instead of refusing a capability you have a tool for.
Prompt-only change; 203 tests pass. (A guardrail mitigates this class of hallucination but can't 100% eliminate it.)
🤖 Generated with Claude Code