@fernado03/zoo-flow 0.5.2 → 0.5.3

package/templates/full/.zoo-flow/evals/routing-cases.md

@@ -1,203 +1,203 @@
-# Routing Eval Cases
-
-Use these cases to check whether the orchestrator chooses the expected workflow.
-
-In every case:
-
-- The user did **not** type a slash command.
-- A free-form request is never self-approving. The orchestrator proposes, then waits.
-- Slash commands, mode names, and executable routing text must not appear in clickable suggestions.
-- Slash commands are optional. The user should never be told to type one to use Zoo Flow.
-
-## Case 1 — Tiny copy change
-
-User:
-"Change the Save button text to Submit."
-
-Expected:
-Recommend the small implementation workflow.
-
-Must not:
-- Route to feature.
-- Read architecture docs by default.
-- Ask the user to type a slash command.
-
-## Case 2 — Unknown crash
-
-User:
-"Checkout randomly crashes after payment. It used to work."
-
-Expected:
-Recommend the diagnosis workflow.
-
-Must:
-- Reproduce before hypothesizing.
-- Present hypotheses before fix.
-
-## Case 3 — New capability
-
-User:
-"Add team invitations with email invites and pending invite states."
-
-Expected:
-Recommend feature planning.
-
-Must:
-- Plan before implementation.
-- Use phase gates.
-
-## Case 4 — Structural cleanup
-
-User:
-"The auth module is getting hard to change. I want to decouple provider-specific logic."
-
-Expected:
-Recommend refactor workflow.
-
-Must:
-- Preserve behavior.
-- Explore architecture candidates before implementation.
-
-## Case 5 — Unknown area
-
-User:
-"I need to change billing but I don't know where that logic lives."
-
-Expected:
-Recommend exploration first.
-
-Must:
-- Produce a map before choosing feature/fix/refactor.
-
-## Case 6 — Known mechanical fix
-
-User:
-"The env var name changed from API_KEY to ZOO_API_KEY. Update the config loader."
-
-Expected:
-Recommend small implementation workflow.
-
-Must not:
-- Route to diagnosis.
-- Route to feature.
-
-## Case 7 — TDD with clear interface
-
-User:
-"Add a slugify helper for article URLs. I want it test-first."
-
-Expected:
-Recommend TDD workflow.
-
-Must:
-- Write the failing test first.
-- Confirm the public interface (input, output, edge cases) is clear before coding.
-
-## Case 8 — Stale documentation
-
-User:
-"The ARCHITECTURE.md file describes a checkout flow we removed last quarter. Bring it in line with the code."
-
-Expected:
-Recommend the documentation update workflow.
-
-Must:
-- Audit code first, then make surgical doc edits.
-- Not rewrite the file wholesale.
-
-## Case 9 — Ready to commit
-
-User:
-"I finished the small tweak. Please commit it and add a journal entry."
-
-Expected:
-Recommend the commit + journal workflow.
-
-Must:
-- Propose a Conventional Commit message and wait for approval before running `git commit` or `git push`.
-
-## Case 10 — Issue triage
-
-User:
-"We have 30 incoming bug reports from the support team. Triage them into the issue tracker."
-
-Expected:
-Recommend the triage workflow.
-
-Must:
-- Ask before publishing, labeling, closing, or making any irreversible tracker change.
-
-## Case 11 — Throwaway design probe
-
-User:
-"I'm not sure if the new search ranking should run inline or in a queue. Can we try both and see?"
-
-Expected:
-Recommend a throwaway prototype.
-
-Must:
-- Keep the work on a prototype branch or `.scratch/prototypes/<slug>/` so it is clearly throwaway.
-- Resolve the design question, not commit to a real implementation.
-
-## Case 12 — Explicit slash command
-
-User:
-"/tweak rename the cancel button to close."
-
-Expected:
-Route immediately. Do not second-guess the explicit command.
-
-Must not:
-- Repropose the workflow as a numbered choice.
-- Treat the explicit command as if approval were still pending.
-
-## Case 13 — Ambiguous "fix" for a known mechanical change
-
-User:
-"Fix the typo in the cancel-button label and update the aria-label to match."
-
-Expected:
-Recommend the small implementation workflow, not diagnosis.
-
-Must:
-- Recognize the cause and target are known.
-- Not run a full diagnosis loop for a one-line copy fix.
-
-## Case — Free-form request must not expose slash commands
-
-User:
-"Change the Save button text to Submit."
-
-Expected:
-Recommend the small implementation workflow in plain language.
-
-Good response:
-"This looks like a small implementation change because the target is known and the risk is low.
-
-1. Make the small implementation change
-2. Explore the area first"
-
-Must not:
-- Say "use `/tweak`" in the user-facing recommendation.
-- Offer `/tweak` as a selectable option.
-- Tell the user to type a slash command.
-
-Allowed:
-- Internally delegate using `/tweak` after the user approves.
-- Mention slash commands only if the user explicitly asks for command syntax.
-
-## Case — Deep inspection must not route to Ask mode
-
-User:
-"Do you think these changes are beneficial or not? Inspect deeply if it affects the system."
-
-Expected:
-Recommend analysis/review through the architecture/inspection workflow.
-
-Delegation target after approval:
-`system-architect`
-
-Must not:
-- Delegate to Ask mode.
-- Delegate to default Architect mode.
-- Use any mode other than `system-architect` or `code-tweaker`.
+# Routing Eval Cases
+
+Use these cases to check whether the orchestrator chooses the expected workflow.
+
+In every case:
+
+- The user did **not** type a slash command.
+- A free-form request is never self-approving. The orchestrator proposes, then waits.
+- Slash commands, mode names, and executable routing text must not appear in clickable suggestions.
+- Slash commands are optional. The user should never be told to type one to use Zoo Flow.
+
+## Case 1 — Tiny copy change
+
+User:
+"Change the Save button text to Submit."
+
+Expected:
+Recommend the small implementation workflow.
+
+Must not:
+- Route to feature.
+- Read architecture docs by default.
+- Ask the user to type a slash command.
+
+## Case 2 — Unknown crash
+
+User:
+"Checkout randomly crashes after payment. It used to work."
+
+Expected:
+Recommend the diagnosis workflow.
+
+Must:
+- Reproduce before hypothesizing.
+- Present hypotheses before fix.
+
+## Case 3 — New capability
+
+User:
+"Add team invitations with email invites and pending invite states."
+
+Expected:
+Recommend feature planning.
+
+Must:
+- Plan before implementation.
+- Use phase gates.
+
+## Case 4 — Structural cleanup
+
+User:
+"The auth module is getting hard to change. I want to decouple provider-specific logic."
+
+Expected:
+Recommend refactor workflow.
+
+Must:
+- Preserve behavior.
+- Explore architecture candidates before implementation.
+
+## Case 5 — Unknown area
+
+User:
+"I need to change billing but I don't know where that logic lives."
+
+Expected:
+Recommend exploration first.
+
+Must:
+- Produce a map before choosing feature/fix/refactor.
+
+## Case 6 — Known mechanical fix
+
+User:
+"The env var name changed from API_KEY to ZOO_API_KEY. Update the config loader."
+
+Expected:
+Recommend small implementation workflow.
+
+Must not:
+- Route to diagnosis.
+- Route to feature.
+
+## Case 7 — TDD with clear interface
+
+User:
+"Add a slugify helper for article URLs. I want it test-first."
+
+Expected:
+Recommend TDD workflow.
+
+Must:
+- Write the failing test first.
+- Confirm the public interface (input, output, edge cases) is clear before coding.
+
+## Case 8 — Stale documentation
+
+User:
+"The ARCHITECTURE.md file describes a checkout flow we removed last quarter. Bring it in line with the code."
+
+Expected:
+Recommend the documentation update workflow.
+
+Must:
+- Audit code first, then make surgical doc edits.
+- Not rewrite the file wholesale.
+
+## Case 9 — Ready to commit
+
+User:
+"I finished the small tweak. Please commit it and add a journal entry."
+
+Expected:
+Recommend the commit + journal workflow.
+
+Must:
+- Propose a Conventional Commit message and wait for approval before running `git commit` or `git push`.
+
+## Case 10 — Issue triage
+
+User:
+"We have 30 incoming bug reports from the support team. Triage them into the issue tracker."
+
+Expected:
+Recommend the triage workflow.
+
+Must:
+- Ask before publishing, labeling, closing, or making any irreversible tracker change.
+
+## Case 11 — Throwaway design probe
+
+User:
+"I'm not sure if the new search ranking should run inline or in a queue. Can we try both and see?"
+
+Expected:
+Recommend a throwaway prototype.
+
+Must:
+- Keep the work on a prototype branch or `.scratch/prototypes/<slug>/` so it is clearly throwaway.
+- Resolve the design question, not commit to a real implementation.
+
+## Case 12 — Explicit slash command
+
+User:
+"/tweak rename the cancel button to close."
+
+Expected:
+Route immediately. Do not second-guess the explicit command.
+
+Must not:
+- Repropose the workflow as a numbered choice.
+- Treat the explicit command as if approval were still pending.
+
+## Case 13 — Ambiguous "fix" for a known mechanical change
+
+User:
+"Fix the typo in the cancel-button label and update the aria-label to match."
+
+Expected:
+Recommend the small implementation workflow, not diagnosis.
+
+Must:
+- Recognize the cause and target are known.
+- Not run a full diagnosis loop for a one-line copy fix.
+
+## Case — Free-form request must not expose slash commands
+
+User:
+"Change the Save button text to Submit."
+
+Expected:
+Recommend the small implementation workflow in plain language.
+
+Good response:
+"This looks like a small implementation change because the target is known and the risk is low.
+
+1. Make the small implementation change
+2. Explore the area first"
+
+Must not:
+- Say "use `/tweak`" in the user-facing recommendation.
+- Offer `/tweak` as a selectable option.
+- Tell the user to type a slash command.
+
+Allowed:
+- Internally delegate using `/tweak` after the user approves.
+- Mention slash commands only if the user explicitly asks for command syntax.
+
+## Case — Deep inspection must not route to Ask mode
+
+User:
+"Do you think these changes are beneficial or not? Inspect deeply if it affects the system."
+
+Expected:
+Recommend analysis/review through the architecture/inspection workflow.
+
+Delegation target after approval:
+`system-architect`
+
+Must not:
+- Delegate to Ask mode.
+- Delegate to default Architect mode.
+- Use any mode other than `system-architect` or `code-tweaker`.