fix(addie): audit and remove always-available tool duplicates from set arrays#3005
Merged
Merged
Conversation
…t arrays Closes #2998 Removes 9 tools from set arrays where they duplicated entries in ALWAYS_AVAILABLE_TOOLS / ALWAYS_AVAILABLE_ADMIN_TOOLS. The duplicates caused Sonnet to hallucinate capability unavailability (e.g. "I can't submit content") when the router didn't select the set whose description mentioned the capability. Changes: - member.tools: remove propose_content, get_my_content, set_outreach_preference - content.tools: remove list_pending_content, approve_content, reject_content - knowledge.tools: remove get_github_issue - admin.tools: remove list_escalations, resolve_escalation - Reword member and content descriptions to not claim ownership of always-available capabilities - Add 7 entries to ALWAYS_AVAILABLE_BLURBS so the correction section covers reclaimed tools - Add tools-array invariant test that unions both always-available arrays Non-breaking: tools remain reachable via ALWAYS_AVAILABLE_TOOLS / ALWAYS_AVAILABLE_ADMIN_TOOLS; removing from set arrays has no functional impact on tool availability for end users. ad-tech-protocol-expert and prompt-engineer reviewed; both approved. Session: https://claude.ai/code/session_01Qyi4Qs9RU6WkxmMxQ3yjrv
Contributor
IPR Policy Agreement RequiredThank you for your contribution! Before we can accept your pull request, you must agree to our Intellectual Property Rights Policy. By making a Contribution, you agree that:
To agree, please comment below with the exact phrase: You can read the full IPR Policy here. I have read the IPR Policy You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot. |
Closed
4 tasks
Contributor
Author
|
I have read the IPR Policy |
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.
Closes #2998
Summary
PR #2979 fixed the specific Addie hallucination where
get_github_issue/draft_github_issue/create_github_issuewere duplicated betweenALWAYS_AVAILABLE_TOOLSand a set'stoolsarray. The same pattern existed in 9 other places. This PR completes the systematic audit.Root cause: When a tool appears in both
ALWAYS_AVAILABLE_TOOLSand a set'stoolsarray, the set's description (by construction) describes that capability. When the set lands in the "Capabilities Not Available" hint, Sonnet reads the description and hallucinates the capability is off — even though the tool is loaded viaALWAYS_AVAILABLE_TOOLS.Changes
server/src/addie/tool-sets.tsmember.tools: removepropose_content,get_my_content,set_outreach_preferencecontent.tools: removelist_pending_content,approve_content,reject_contentknowledge.tools: removeget_github_issueadmin.tools: removelist_escalations,resolve_escalationmemberandcontentdescriptions to not claim ownership of always-available capabilitiesALWAYS_AVAILABLE_BLURBSwith 7 new entries:get_escalation_status,propose_content,get_my_content,list_pending_content,approve_content,reject_content,set_outreach_preferenceserver/tests/unit/addie/tool-sets.test.tsALWAYS_AVAILABLE_TOOLS∪ALWAYS_AVAILABLE_ADMIN_TOOLSappears in any set'stoolsarray. Fails loudly with the exact set and tool name. This replaces the hand-auditing the issue requested as a "spreadsheet" — it's zero-maintenance and catches future regressions automatically.Non-breaking justification
All removed tools remain reachable in every conversation via
ALWAYS_AVAILABLE_TOOLS/ALWAYS_AVAILABLE_ADMIN_TOOLS. Removing from set arrays is strictly a deduplication with no functional change to what tools Sonnet can call.getToolsForSetsbuilds the final tool list as aSet, so ALWAYS_AVAILABLE already guarantees these tools are loaded regardless of routing.Expert consensus
prompt-engineer and adtech-product-expert reviewed; both approved. Key finding:
propose_content/get_my_contentis the highest-priority fix (content submission is the primary member workflow and the hallucination destroys trust at the moment a member is most invested).Session: https://claude.ai/code/session_01Qyi4Qs9RU6WkxmMxQ3yjrv
Generated by Claude Code