intentdna 1.5.8 → 1.5.10

package/dist/templates/flutter-rewrite.dna.yaml

@@ -316,6 +316,12 @@ workflows:
         role: surgeon
         depends_on: [investigate]
         description: "Fix identified issues. Max 5 files per round."
+        max_attempts: 3
+        on_fail: handoff
+        handoff_to: investigate
+        max_handoffs: 3
+        on_handoff_exhausted: skip
+        blocked_items_path: "docs/behavior/blocked_items.md"
         checkpoints:
           - assert: clean_working_tree
             message: "Commit all changes before proceeding"
@@ -337,9 +343,26 @@ workflows:
           produces:
             - type: git_commit
               description: "Rescue round commit"
-      - id: review
+      - id: progress_check
         role: investigator
         depends_on: [fix]
+        description: "Check test progress after surgeon's fix."
+        prompt: |
+          Run tests. Compare green count with previous round.
+          If green count increased: PROGRESS — proceed to review.
+          If green count unchanged or decreased: NO_PROGRESS — record what surgeon tried
+          and why it didn't work, for the experience chain.
+        handoff:
+          consumes:
+            - type: git_commit
+              from: fix
+              description: "Fix commit from surgeon"
+          produces:
+            - type: summary
+              description: "Progress check result (PROGRESS or NO_PROGRESS)"
+      - id: review
+        role: investigator
+        depends_on: [progress_check]
         description: "Read-only review of surgeon's changes."
         prompt: |
           Review surgeon's git diff (read-only, do NOT modify any files):
@@ -352,6 +375,9 @@ workflows:
           Verdict: REQUEST_CHANGES → describe specific problems. Next round's investigate step will include this feedback.
         handoff:
           consumes:
+            - type: summary
+              from: progress_check
+              description: "Progress check result"
             - type: git_commit
               from: fix
               description: "Committed fix from surgeon"

package/dist/templates/persistent-executor.dna.yaml

@@ -0,0 +1,144 @@
+version: "0.1.0"
+id: template_persistent_executor
+name: Persistent Executor (PRD-Driven)
+type: project
+namespace: pe
+
+cascade:
+  inherits: ["species:default"]
+  priority: 100
+
+genes:
+  prd_driven:
+    description: Every task must trace back to explicit acceptance criteria
+    codons:
+      - type: threshold
+        condition: "acceptance_criteria_defined == true"
+        action: block
+      - type: attract
+        target: reference_acceptance_criteria
+      - type: repel
+        target: work_without_spec
+
+  verify_before_done:
+    description: Never claim completion without passing acceptance criteria
+    codons:
+      - type: threshold
+        condition: "all_criteria_verified == true"
+        action: block
+      - type: attract
+        target: run_verification_before_reporting
+      - type: repel
+        target: claim_done_without_evidence
+
+  no_partial_completion:
+    description: Either fully complete a story or explicitly report what remains
+    codons:
+      - type: threshold
+        condition: "story_fully_complete == true"
+        action: escalate
+      - type: repel
+        target: silent_partial_delivery
+
+  small_increments:
+    description: One story per commit, verifiable increments
+    role: workflow_hint
+    codons:
+      - type: attract
+        target: one_story_one_commit
+      - type: repel
+        target: batch_multiple_stories
+      - type: attract
+        target: each_commit_passes_tests
+
+contexts: {}
+
+roles:
+  executor:
+    description: Implements stories from the PRD. Can modify code.
+    tool_permissions:
+      allow: [Read, Edit, Write, Grep, Glob, Bash]
+    scope:
+      read: ["**/*"]
+      write: ["src/**", "lib/**", "test/**"]
+    instructions:
+      - Pick the highest priority incomplete story
+      - Read acceptance criteria before starting
+      - Implement in small verifiable steps
+      - Run tests after each change
+      - Commit after each story completion
+      - "Commit message: feat($STORY): what was implemented"
+      - Do not move to next story until current one is verified
+
+  verifier:
+    description: Independently verifies story completion. Read-only + can run tests.
+    tool_permissions:
+      allow: [Read, Grep, Glob, Bash]
+      deny: [Edit, Write, NotebookEdit]
+    scope:
+      read: ["**/*"]
+      write: []
+    instructions:
+      - Read the acceptance criteria for the completed story
+      - Run tests independently (do not trust executor's output)
+      - Check each criterion individually
+      - "Mark each: VERIFIED / PARTIAL / MISSING"
+      - "Verdict: PASS (all verified) or FAIL (list what's missing)"
+      - On FAIL the executor must fix before proceeding
+
+workflows:
+  persistent-execution:
+    name: Persistent Execution
+    description: "PRD-driven story execution: pick, implement, verify, repeat until all done"
+    steps:
+      - id: pick_story
+        role: executor
+        description: "Select next story from PRD by priority."
+        prompt: |
+          Read the PRD or task list.
+          Select the highest priority incomplete story.
+          Confirm: acceptance criteria are clear and testable.
+          If criteria are vague, ask for clarification before starting.
+          Output: story ID, title, acceptance criteria.
+
+      - id: implement
+        role: executor
+        depends_on: [pick_story]
+        description: "Implement the selected story."
+        prompt: |
+          Implement the story. Rules:
+          - Read acceptance criteria carefully
+          - Small steps: change → test → verify
+          - Maximum scope: what the story requires, nothing more
+          - Run tests after each meaningful change
+          - If stuck for 3 attempts on the same issue: STOP and report blocker
+          - Commit when story implementation is complete
+          - Do not start the next story
+
+      - id: verify
+        role: verifier
+        depends_on: [implement]
+        description: "Independent verification of story completion."
+        prompt: |
+          Verify the completed story independently:
+          1. Read acceptance criteria
+          2. Run all relevant tests (do not trust executor's results)
+          3. Check each criterion: VERIFIED / PARTIAL / MISSING
+          4. Run full test suite for regression check
+
+          PASS: All criteria verified → story complete, ready for next
+          FAIL: List what's missing → executor must fix
+
+      - id: report
+        role: verifier
+        depends_on: [verify]
+        description: "Report progress and determine next action."
+        prompt: |
+          Summary:
+          - Story: [ID] [title]
+          - Criteria: N/M verified
+          - Tests: X passing, Y failing
+          - Verdict: COMPLETE or NEEDS_FIX
+
+          If COMPLETE: "Story done. Pick next story to continue."
+          If NEEDS_FIX: "Story incomplete. Issues: [list]. Fix before proceeding."

package/dist/templates/qa-loop.dna.yaml

@@ -0,0 +1,147 @@
+version: "0.1.0"
+id: template_qa_loop
+name: QA Loop (Test-Fix Cycle)
+type: project
+namespace: ql
+
+cascade:
+  inherits: ["species:default"]
+  priority: 100
+
+genes:
+  convergence_protection:
+    description: Stop if no progress after 2 consecutive rounds
+    codons:
+      - type: threshold
+        condition: "consecutive_no_progress_rounds <= 2"
+        action: block
+      - type: attract
+        target: track_test_progress_per_round
+      - type: repel
+        target: infinite_retry_loop
+
+  separate_test_fix:
+    description: Tester and fixer must be separate roles
+    codons:
+      - type: threshold
+        condition: "tester_is_not_fixer == true"
+        action: block
+      - type: repel
+        target: self_test_own_fix
+
+  evidence_based_diagnosis:
+    description: Diagnose from test output, not guessing
+    role: workflow_hint
+    codons:
+      - type: attract
+        target: read_error_output_first
+      - type: repel
+        target: guess_without_running_tests
+      - type: attract
+        target: categorize_failures_by_type
+
+  regression_awareness:
+    description: Fixes must not introduce new failures
+    codons:
+      - type: threshold
+        condition: "new_failures_introduced == 0"
+        action: escalate
+      - type: attract
+        target: run_full_suite_after_fix
+
+contexts: {}
+
+roles:
+  tester:
+    description: Runs tests, diagnoses failures, categorizes issues. Read-only.
+    tool_permissions:
+      allow: [Read, Grep, Glob, Bash]
+      deny: [Edit, Write, NotebookEdit]
+    scope:
+      read: ["**/*"]
+      write: []
+    instructions:
+      - Run the full test suite
+      - Categorize failures by type (compile, logic, integration, flaky)
+      - For each failure trace the root cause
+      - Prioritize by severity
+      - Compare results with previous round — track progress
+      - Do NOT suggest code changes
+
+  fixer:
+    description: Fixes issues identified by tester. Can modify code.
+    tool_permissions:
+      allow: [Read, Edit, Write, Grep, Glob, Bash]
+    scope:
+      read: ["**/*"]
+      write: ["src/**", "lib/**", "test/**"]
+    instructions:
+      - Fix issues in priority order (compile first, then logic, then integration)
+      - One fix at a time, run tests after each
+      - If a fix introduces new failures, revert immediately
+      - Maximum 5 files per round
+      - "Same issue failing 3 times → STOP, report as blocked"
+      - Commit after each successful fix batch
+
+workflows:
+  qa-cycle:
+    name: QA Cycle
+    description: "Test-fix loop: test, diagnose, fix, retest. Max 5 rounds with convergence protection."
+    max_rounds: 5
+    convergence_rule: "2 consecutive rounds with 0 test progress → STOP"
+    steps:
+      - id: test
+        role: tester
+        description: "Run tests and categorize failures."
+        prompt: |
+          Run the test suite. Record results:
+          - Total tests, passing, failing, skipped
+          - If round > 1: compare with previous round results
+          - Categorize each failure:
+            COMPILE: syntax/import errors — fix first
+            LOGIC: assertion failures — fix after compile
+            INTEGRATION: cross-module issues — fix after logic
+            FLAKY: intermittent failures — deprioritize
+          - If 0 failures: "All tests passing. QA complete."
+          - If round > 1 and no progress: warn "no improvement"
+
+      - id: diagnose
+        role: tester
+        depends_on: [test]
+        description: "Root cause analysis for top failures."
+        prompt: |
+          For the top 3 failures (by severity):
+          1. Read the error output carefully
+          2. Trace to the root cause in source code
+          3. Identify the minimal fix needed
+          4. Check if failures are related (shared root cause)
+          Output: prioritized fix list with file paths and line numbers.
+
+      - id: fix
+        role: fixer
+        depends_on: [diagnose]
+        description: "Apply fixes for diagnosed issues."
+        prompt: |
+          Fix the diagnosed issues. Rules:
+          - Follow the priority order from diagnosis
+          - One fix at a time → run tests → verify
+          - If fix causes new failures: revert and try different approach
+          - Maximum 5 files per round
+          - Same issue failed 3 times → STOP and report as blocked
+          - Commit: "fix($TYPE): description of what was fixed"
+
+      - id: retest
+        role: tester
+        depends_on: [fix]
+        description: "Verify fixes and check for regressions."
+        prompt: |
+          Run the full test suite again:
+          1. Are the targeted failures fixed?
+          2. Any new regressions introduced?
+          3. Test delta: +N fixed, -M new failures
+
+          Progress check:
+          - If all tests pass: "QA complete. All green."
+          - If progress made (fewer failures): "Round N: +X fixed. Continue."
+          - If no progress (same or more failures): "No progress. [2 rounds → STOP]"
+          - If blocked items exist: "Blocked on: [list]. Human intervention needed."

package/dist/templates/requirements-gate.dna.yaml

@@ -0,0 +1,116 @@
+# requirements-gate: 需求清晰度门控
+# 在需求明确（>=80%）之前阻止实现，每轮只问一个问题
+# 适用: 任何需要"先澄清需求再动手"的项目
+
+version: "0.1.0"
+id: template_requirements_gate
+name: Requirements Clarity Gate
+type: project
+namespace: rg
+
+cascade:
+  inherits: ["species:default"]
+  priority: 100
+
+genes:
+  clarity_threshold:
+    description: Block implementation until requirements are sufficiently clear (>=80%)
+    codons:
+      - type: threshold
+        condition: "requirements_clarity_score >= 80"
+        action: block
+      - type: attract
+        target: clarify_before_building
+      - type: repel
+        target: implement_with_ambiguity
+
+  one_question_per_round:
+    description: Ask one focused question at a time to reduce cognitive load
+    role: workflow_hint
+    codons:
+      - type: attract
+        target: single_focused_question
+      - type: repel
+        target: question_dump
+
+  target_weakest_dimension:
+    description: Always probe the least clear dimension first
+    role: workflow_hint
+    codons:
+      - type: attract
+        target: identify_weakest_area
+      - type: repel
+        target: redundant_clarification
+
+  structured_output:
+    description: Requirements must be structured and actionable
+    codons:
+      - type: attract
+        target: testable_acceptance_criteria
+      - type: repel
+        target: vague_requirements
+      - type: attract
+        target: explicit_constraints
+
+contexts: {}
+
+roles:
+  interviewer:
+    description: Evaluates requirements clarity and asks clarifying questions. Read-only.
+    tool_permissions:
+      allow: [Read, Grep, Glob, Bash]
+      deny: [Edit, Write, NotebookEdit]
+    scope:
+      read: ["**/*"]
+      write: []
+    instructions:
+      - Evaluate requirements across 4 dimensions (Goal, Constraints, Criteria, Context)
+      - Score each dimension 0-100
+      - If average < 80 output the most critical question targeting the weakest dimension
+      - If average >= 80 output structured requirements document
+      - One question per round — do not dump multiple questions
+      - "80% threshold rationale: below this, >1/5 dimensions are underspecified"
+
+workflows:
+  requirements-gate:
+    name: Requirements Gate
+    description: "Evaluate requirements clarity, block if ambiguous, proceed if clear"
+    steps:
+      - id: clarity_check
+        role: interviewer
+        description: "Score requirements across 4 dimensions."
+        prompt: |
+          Evaluate the given requirements across 4 dimensions:
+
+          1. Goal (0-100): Is the desired outcome clear and measurable?
+          2. Constraints (0-100): Are boundaries, limitations, and non-goals defined?
+          3. Criteria (0-100): Are acceptance criteria specific and testable?
+          4. Context (0-100): Is the technical/business context sufficient?
+
+          Calculate average score.
+
+          If average >= 80:
+            Output: structured requirements document with all 4 dimensions filled.
+            Status: PROCEED
+
+          If average < 80:
+            Identify the weakest dimension.
+            Output: one specific clarifying question targeting that dimension.
+            Status: BLOCK — "Requirements clarity at N%. Need clarification on: [dimension]"
+
+      - id: proceed_or_block
+        role: interviewer
+        depends_on: [clarity_check]
+        description: "Gate decision based on clarity score."
+        prompt: |
+          Based on clarity check results:
+
+          If PROCEED: Format the requirements as a structured spec:
+          - Goal: [what]
+          - Constraints: [boundaries]
+          - Acceptance Criteria: [testable conditions]
+          - Context: [technical/business background]
+          - Out of Scope: [explicit exclusions]
+
+          If BLOCK: Summarize what's missing and the question to ask.
+          Do NOT proceed with implementation planning until clarity >= 80%.

package/dist/templates/research-orchestration.dna.yaml

@@ -0,0 +1,145 @@
+# research-orchestration: 并行假设驱动研究
+# 适用: 需要结构化调研、多角度验证、有据可查结论的研究任务
+
+version: "0.1.0"
+id: template_research_orchestration
+name: Research Orchestration (Parallel Hypotheses)
+type: project
+namespace: ro
+
+cascade:
+  inherits: ["species:default"]
+  priority: 100
+
+genes:
+  hypothesis_driven:
+    description: Research must start with explicit hypotheses before investigation
+    codons:
+      - type: threshold
+        condition: "hypotheses_defined == true"
+        action: block
+      - type: attract
+        target: state_hypothesis_before_investigating
+      - type: repel
+        target: undirected_exploration
+
+  cross_verify:
+    description: Findings must be cross-verified by independent researchers
+    role: workflow_hint
+    codons:
+      - type: attract
+        target: independent_verification
+      - type: repel
+        target: single_source_conclusion
+
+  evidence_graded:
+    description: Grade evidence quality and distinguish fact from inference
+    role: workflow_hint
+    codons:
+      - type: attract
+        target: grade_source_reliability
+      - type: attract
+        target: separate_fact_from_interpretation
+      - type: repel
+        target: treat_all_evidence_equally
+
+  structured_synthesis:
+    description: Synthesis must address all hypotheses with evidence
+    codons:
+      - type: threshold
+        condition: "all_hypotheses_addressed == true"
+        action: escalate
+      - type: attract
+        target: evidence_based_conclusions
+
+contexts: {}
+
+roles:
+  lead_researcher:
+    description: Decomposes questions, synthesizes findings. Read-only.
+    tool_permissions:
+      allow: [Read, Grep, Glob, Bash]
+      deny: [Edit, Write, NotebookEdit]
+    scope:
+      read: ["**/*"]
+      write: []
+    instructions:
+      - Decompose the research question into sub-questions
+      - Formulate hypotheses for each sub-question
+      - After investigation synthesize findings across all researchers
+      - Grade overall confidence in conclusions
+      - Identify remaining unknowns and next steps
+
+  researcher:
+    description: Investigates specific sub-questions. Read-only with analysis tools.
+    tool_permissions:
+      allow: [Read, Grep, Glob, Bash]
+      deny: [Edit, Write, NotebookEdit]
+    scope:
+      read: ["**/*"]
+      write: []
+    instructions:
+      - Focus on assigned sub-question only
+      - State hypothesis before investigating
+      - Gather evidence from code, docs, logs, tests
+      - Grade each piece of evidence (strong/moderate/weak)
+      - Report findings with evidence quality ratings
+      - Flag uncertainties and assumptions explicitly
+
+workflows:
+  research:
+    name: Research
+    description: "Parallel hypothesis research: decompose, investigate, verify, synthesize"
+    steps:
+      - id: decompose
+        role: lead_researcher
+        description: "Break research question into sub-questions with hypotheses."
+        prompt: |
+          Analyze the research question:
+          1. Break into 2-4 independent sub-questions
+          2. For each sub-question:
+             - Hypothesis: what do we expect to find?
+             - Key sources: where to look (code, docs, logs, external)
+             - Success criteria: what constitutes a sufficient answer?
+          3. Identify dependencies between sub-questions
+          4. Assign investigation priority
+
+      - id: investigate
+        role: researcher
+        depends_on: [decompose]
+        description: "Investigate sub-questions in parallel."
+        prompt: |
+          Investigate assigned sub-question:
+          1. State the hypothesis clearly
+          2. Gather evidence from identified sources
+          3. For each evidence item:
+             - Source: where found
+             - Quality: strong / moderate / weak
+             - Supports or contradicts hypothesis?
+          4. Preliminary conclusion with confidence level
+          5. List remaining unknowns
+
+      - id: verify
+        role: lead_researcher
+        depends_on: [investigate]
+        description: "Cross-verify findings across researchers."
+        prompt: |
+          Cross-verification:
+          1. Do different researchers' findings contradict?
+          2. Are there gaps — sub-questions not fully answered?
+          3. Check: is the evidence quality sufficient for conclusions?
+          4. If contradictions exist: identify the most reliable source
+          5. If gaps exist: note what additional investigation is needed
+
+      - id: synthesize
+        role: lead_researcher
+        depends_on: [verify]
+        description: "Synthesize all findings into conclusions."
+        prompt: |
+          Final synthesis:
+          1. For each original sub-question: answer with evidence
+          2. For each hypothesis: confirmed / refuted / inconclusive
+          3. Overall conclusion with confidence level (high/medium/low)
+          4. Key findings (bullet points)
+          5. Remaining unknowns
+          6. Recommended next steps (if any)