testops-copilot/specs/features/react-loop.feature.yaml at main · rayalon1984/testops-copilot · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
feature: react-loop
name: "Agentic ReAct Loop"
version: "1.0.0"
status: shipped
since: "v2.9.0"
owner: AI_ARCHITECT
category: ai-tools
spec_source: AI_TOOLS.md

capabilities:
  - id: react.loop
    name: "ReAct Loop Engine"
    version: "1.0.0"
    description: "Reason-Act-Observe loop with safety bounds"
    files:
      - backend/src/services/ai/AIChatService.ts
    assertions:
      - id: react.loop.max-iterations
        description: "Loop terminates at MAX_REACT_ITERATIONS (8) with error event"
        type: invariant
      - id: react.loop.max-tool-calls
        description: "Tool execution stops at MAX_TOOL_CALLS (5) per request"
        type: invariant
      - id: react.loop.tool-not-found-graceful
        description: "Unknown tool name adds error message to history, loop continues (never throws)"
        type: invariant
      - id: react.loop.no-tool-calls-is-answer
        description: "Response with no tool calls is treated as final answer"
        type: invariant

  - id: react.tools
    name: "Tool Execution"
    version: "1.0.0"
    description: "Tool dispatch with role enforcement and autonomy tiers"
    files:
      - backend/src/services/ai/AIChatService.ts
    assertions:
      - id: react.tools.role-enforcement
        description: "Tool requiring higher role than user returns access denied (tool NOT executed)"
        type: invariant
      - id: react.tools.tier1-auto-execute
        description: "Tier 1 tools auto-execute and emit 'autonomous_action' SSE event"
        type: invariant
      - id: react.tools.tier3-pauses-loop
        description: "Tier 3 tools create pending action and pause loop with confirmation_request"
        type: invariant

  - id: react.streaming
    name: "SSE Streaming"
    version: "1.0.0"
    description: "Server-Sent Events for real-time client communication"
    files:
      - backend/src/services/ai/AIChatService.ts
    assertions:
      - id: react.streaming.event-types
        description: "SSE events include: thinking, tool_start, tool_result, answer_chunk, answer, done, error"
        type: contract
      - id: react.streaming.ui-context-injection
        description: "If uiContext exists, it is prepended to user message as [UI Context: ...]"
        type: behavioral
      - id: react.streaming.timestamps
        description: "Each SSE event includes a timestamp field"
        type: invariant

  - id: react.tools-registry
    name: "Tool Registry"
    version: "1.0.0"
    description: "Registry of available AI tools (read and write)"
    files:
      - backend/src/services/ai/tools/index.ts
    assertions:
      - id: react.registry.read-tools-count
        description: "7 read-only tools registered: jira_search, jira_get, github_get_commit, github_get_pr, confluence_search, jenkins_get_status, dashboard_metrics"
        type: invariant
      - id: react.registry.write-tools-count
        description: "6+ write tools registered with requiresConfirmation flag"
        type: invariant

acceptance_criteria:
  - id: AC-RL-1
    title: "ReAct Loop Safety"
    gherkin: |
      GIVEN the AI copilot is processing a user message
      WHEN the ReAct loop executes tool calls
      THEN it never exceeds 5 tool calls or 8 iterations
      AND unknown tools are handled gracefully without crashes
      AND role-insufficient tool calls are blocked
    maps_to:
      - react.loop.max-iterations
      - react.loop.max-tool-calls
      - react.loop.tool-not-found-graceful
      - react.tools.role-enforcement

changelog:
  - version: "1.0.0"
    date: "2026-02-21"
    product_version: "v3.0.0"
    changes:
      - "Feature spec created: ReAct loop, tool execution, SSE streaming, tool registry"