fix(mode-efficiency): facilitator Phase 0 discipline + tool-designer output-first gate#24
Open
colombod wants to merge 1 commit into
Open
Conversation
…output-first gate
Three problems observed across real sessions and confirmed via automated eval:
1. Facilitator opened with a multi-question batch instead of one WHY question.
Users had to steer 3-5 times per session to get a single focused question.
2. On delegate:error, the mode silently fell back to inline Phase 0 instead
of stopping and reporting the failure.
3. Tool-designer wrote an output mockup to disk but never showed it in its
response — the user had to ask to see it.
Changes
-------
agents/context-intelligence-design-facilitator.md
- Opening move rewritten: one sentence, one question mark, stop.
- Banned phrasing enumerated: questionnaire-style openers ('Why do you need...',
'What problem are you solving?') replaced with goal-anchored 'what will you
do with these findings?'
- Channel B (source-code exploration) added alongside Channel A (event-stream)
for signal discovery — finds instrumentation gaps that Channel A misses.
- Proactive phase-transition offer added: facilitator calls the handoff
explicitly rather than waiting for the user to prompt.
modes/context-intelligence.md
- delegate:error handler upgraded from a paragraph to a numbered 3-step
hard stop: (1) name the failure in one sentence, (2) stop completely,
(3) wait. Explicit: 'This rule overrides everything else in this document.'
agents/context-intelligence-tool-designer.md
- Output-first gate tightened: table must appear inline in the agent's
response (not just written to disk). Required table format documented.
'never describe it, show it.'
skills/workflow-pattern-analysis/SKILL.md
- Phase gate added: checks domain-concepts.md before any analysis begins.
Delegates to facilitator if Phase 0 is incomplete.
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.
What this fixes
Three problems observed across real user sessions and confirmed by an automated eval harness (scenarios S-α, S-δ, S-ε):
1. Facilitator opened with a questionnaire, not a WHY question (S-α)
The facilitator sent 3-5 questions in a batch. Users had to steer multiple times per session before a single focused WHY question emerged. In the eval, S-α was failing with
corrective_cap_exceededconsistently. After this change, S-α accepts — the facilitator sends one open question and waits.2. On delegate:error, the mode fell back to inline Phase 0 silently (S-δ)
When the facilitator failed to spawn, the mode continued with the investigation rather than stopping and reporting the failure. Users had no visibility into what went wrong. The mode text had a recovery hint but it was a soft suggestion buried in a paragraph. After this change, it is a numbered 3-step hard stop that explicitly overrides everything else.
3. Tool-designer wrote the output mockup to disk but never showed it (S-ε)
The Phase 2 output-first gate wrote
output-example.mdbut did not require the table to appear in the agent's response. The user couldn't evaluate the design without separately asking to see the file. After this change, the table must appear inline in the response.Files changed
agents/context-intelligence-design-facilitator.md— one-question opening move, Channel B (source-code exploration) for signal discovery, proactive phase-transition offermodes/context-intelligence.md— delegate:error becomes a 3-step numbered hard stop with "This rule overrides everything else in this document"agents/context-intelligence-tool-designer.md— output-first gate requires table visible in response, not just written to diskskills/workflow-pattern-analysis/SKILL.md— Phase gate added: checks domain-concepts.md before any analysis, delegates to facilitator if Phase 0 incompleteGenerated with Amplifier