File size: 3,444 Bytes
80fa878 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 | # WORKFLOW.md
AGENT_WORKFLOW uses nine gates. A task cannot be considered done unless each gate is completed or explicitly justified.
## Gate 1 — Understand
- Define the goal.
- Define non-goals.
- Define expected output.
- Identify project and task ID.
- Confirm constraints, data sensitivity and deployment target.
## Gate 2 — Search Existing Code
Use `rg`, `find`, `ls`, `tree` or equivalent commands.
Search for:
- existing functions
- schemas
- tests
- conventions
- configs
- scripts
- similar reports or ADRs
Document commands used and files found.
## Gate 3 — Reuse Decision
Choose exactly one:
- Reuse existing function.
- Modify existing function.
- Build small module.
- Stop because unnecessary.
The decision must include a reason. Building is allowed only when reuse or modification is not appropriate.
## Gate 4 — Minimal Design
Write a short design:
- files allowed to edit
- behavior change
- tests to run
- known risks
- rollback path if relevant
No global refactor without explicit task scope.
## Gate 5 — Implement Only Assigned Files
- Keep strict scope.
- Do not edit unrelated files.
- Do not perform global formatting unless requested.
- Do not rename public APIs unless the task requires it.
- Keep modules small and readable.
## Gate 6 — Tests
Run relevant tests:
- unit tests
- integration tests
- smoke tests
- lint/type checks when available
Document exact commands. If tests cannot run, explain why and what should be run next.
## Gate 7 — Anti-Leakage / Security Check
For temporal data:
- verify `available_at <= prediction_time`
- verify no future data enters features
- verify labels are not used as prediction features
- verify train/test split is temporal when needed
- verify backtests are walk-forward
For security:
- no secrets
- no tokens in logs
- no `.env` committed
- no direct port exposure unless justified
- no complete article/content storage without validation
Block the task if leakage or secret exposure is detected.
## Gate 8 — Task Report + Context Compression
Write `reports/task_reports/TASK_ID.md` with:
- goal
- non-goals
- summary
- files changed
- commands run
- tests
- reuse decision
- risks
- security/leakage check
- remaining work
- next recommended task
Update `PROJECT_STATE.md` with only durable information: decisions, changed files, current risks, next tasks and recent reports.
## Gate 9 — Review + Merge
Reviewer checklist:
- scope respected
- reuse decision documented
- tests pass or failure is understood
- no secrets
- no leakage
- docs updated
- task report complete
- project state updated
Merge is blocked if tests, leakage, security or scope checks fail.
## Mermaid Workflow
```mermaid
flowchart TD
A[Gate 1 Understand] --> B[Gate 2 Search Existing Code]
B --> C[Gate 3 Reuse Decision]
C -->|reuse| D[Gate 4 Minimal Design]
C -->|modify| D
C -->|build small module| D
C -->|stop unnecessary| R[Report Stop Decision]
D --> E[Gate 5 Implement Assigned Files]
E --> F[Gate 6 Tests]
F --> G[Gate 7 Leakage and Security Check]
G -->|pass| H[Gate 8 Task Report and Context Compression]
G -->|fail| X[Block and Fix]
X --> E
H --> I[Gate 9 Review and Merge]
I -->|approved| J[Merge]
I -->|changes requested| E
```
|