---
name: biz-impact-analysis
description: |
  Use when code changes are complete and need business impact assessment.
  TRIGGER when: impl-verify passes (automatic for P0/P1), or manually via
  /biz-impact-analysis. DO NOT use for pre-implementation requirement analysis
  (use ecw:domain-collab instead).
---

# Business Impact Analysis

After code changes are complete, dispatch the `biz-impact-analysis` agent to analyze the impact of changes on business processes.

**Mode switch**: Update session-state.md MODE marker to `verification`.

**Announce at start:** "Using ecw:biz-impact-analysis to assess business impact of code changes."

**Output language**: Read `ecw.yml` → `project.output_language`. Pass to dispatched agent prompt. Report headings and labels follow this language.

## Trigger

- **Manual**: `/biz-impact-analysis` — Analyze all changes on current branch vs master
- **Manual (specify range)**: `/biz-impact-analysis HEAD~3` — Analyze last N commits
- **Automatic**: Appended automatically after ecw:impl-verify completes

## Flow

1. **Determine Diff Range** — No args: use `git diff master...HEAD`; with args: use `git diff {args}` to get changed file list
2. **Coordinator Preprocessing** — Execute before Agent dispatch:
   1. Run `git diff --stat {diff_range}` to get change statistics
   2. Run `git diff --name-only {diff_range}` to get file list
   3. Read `ecw-path-mappings.md`, map file list to domains
   4. Fill above results into Agent prompt, replacing full diff

> **Knowledge file robustness**: If `ecw-path-mappings.md` is missing, pass the raw file list to the Agent without domain mapping. The Agent will use path-based heuristic grouping and note `[Warning: path mappings not found, domain identification is heuristic]` in the report.
3. **Dispatch biz-impact-analysis agent** (`model: opus`, default from `models.defaults.analysis`; configurable via ecw.yml — business impact analysis is the final safety net; missed impact goes straight to production incidents) — Pass in preprocessed results, await impact analysis report
4. **Return value validation**: Verify the agent's report contains required sections ("Analysis Coverage", "Change Summary", "Direct Impact"). If the report is missing critical sections:
   - Log to Ledger: `[FAILED: biz-impact-analysis, reason: incomplete report]`
   - Retry once with the same model
   - If retry also fails: output the partial report as-is with `[degraded: incomplete analysis]` header, and warn user that manual impact review may be needed
5. **Auto-backfill knowledge base** — If the agent's report contains "Unregistered Cross-Domain Calls", execute the knowledge backfill procedure (see [Knowledge Auto-Backfill](#knowledge-auto-backfill) below)
6. **Present analysis report** — Output the agent's formatted report directly; append backfill summary if any calls were added

## Agent Dispatch Prompt Template

When dispatching the biz-impact-analysis agent, use the following prompt structure:

```
Please analyze the business impact of the following code changes.

## Diff Range

{diff_range}

## Changed File Summary (Coordinator Preprocessed Results)

{git_diff_stat_output}

## Domain Identification Results

{file_to_domain_mapping}

## Instructions

Execute your 5-step analysis process.
Note: Full diff content has been preprocessed by Coordinator, providing file list and domain identification.
In Step 1, only execute `git diff {diff_range} -- {file_path}` for files that need method signature change inspection.
In Step 3 incremental scan, only read specific change content for files matching scan_patterns.
Do not execute `git diff {diff_range}` for full change content on all files.

Please output the impact analysis report in Chinese.
```

## Argument Parsing Rules

| Input | Diff Command |
|-------|-------------|
| `/biz-impact-analysis` | `git diff master...HEAD` |
| `/biz-impact-analysis HEAD~3` | `git diff HEAD~3...HEAD` |
| `/biz-impact-analysis abc123` | `git diff abc123...HEAD` |
| `/biz-impact-analysis abc123 def456` | `git diff abc123...def456` |

## Knowledge Auto-Backfill

When the agent's report flags "Unregistered Cross-Domain Calls", automatically backfill them into the knowledge base instead of relying on manual follow-up.

### Procedure

1. **Extract unregistered calls** — Parse the agent's report for all entries under "Unregistered Cross-Domain Calls" (or equivalent section). Each entry typically contains: caller domain, callee domain, call method (RPC/HTTP/MQ), caller class/method, callee class/method.

2. **Read existing matrix** — Read `.claude/knowledge/common/cross-domain-calls.md` (path from ecw.yml `paths.knowledge_common`).
   - If the file does not exist, skip backfill and output `[Warning: cross-domain-calls.md not found, skipping auto-backfill. Please create the file and backfill manually.]`

3. **Deduplicate** — For each unregistered call, check if an equivalent entry already exists in the matrix (match on caller domain + callee domain + call method + caller class). Skip entries that already exist.

4. **Append new entries** — Append only genuinely new calls to the end of the matrix table, preserving the existing table format. Use Edit tool (not Write) to append rows.
   - **Conservative strategy**: Only add entries. Never modify or delete existing entries.
   - Each appended row should include a trailing comment: `<!-- auto-backfilled by biz-impact-analysis {date} -->`

5. **Report backfill result** — Append the following summary to the end of the analysis report output:

```
---
### 知识库自动回填

已自动回填 {N} 条跨域调用到 `cross-domain-calls.md`:
{list of added entries, one per line}

> 回填策略: 仅追加确认的新调用，不修改/删除现有条目。如需调整请手动编辑。
```

If N=0 (all flagged calls already existed), output instead:
```
---
### 知识库自动回填

报告标记的未注册调用均已存在于知识库中，无需回填。
```

### Edge Cases

| Scenario | Handling |
|----------|---------|
| `cross-domain-calls.md` not found | Skip backfill, warn in report, suggest manual creation |
| `cross-domain-calls.md` has no table header | Skip backfill, warn: `[Warning: cross-domain-calls.md format unrecognized, skipping auto-backfill]` |
| Agent report has no "Unregistered Cross-Domain Calls" section | No backfill needed, skip silently |
| Agent report parsing ambiguous | Skip backfill for ambiguous entries, only backfill clearly parseable ones, note skipped count in summary |

## Integration with impl-verify

When `ecw:impl-verify` completes:

1. impl-verify completes code correctness + quality verification (zero must-fix)
2. Dispatch biz-impact-analysis agent based on the same diff range
3. Output business impact analysis report

**Mandatory for P0/P1 changes**; suggested for P2+. The routing chain output by ecw:risk-classifier Phase 1 already includes `ecw:impl-verify + ecw:biz-impact-analysis`, and Phase 1 adds them to TaskCreate todo list.

## Integration with Phase 3

After biz-impact-analysis report is output:

> **Knowledge utilization tracking**: If ecw.yml `paths.knowledge_root` exists (ECW-ready session):
> **Immediately invoke** `ecw:knowledge-track`. Do NOT output a suggestion or wait for user response.
> Mark the biz-impact-analysis Task as complete; then proceed to the Auto-Continue Rule below.

> **CRITICAL — Auto-Continue Rule**: Read risk level from session-state.md, update `Next` field, then:
> - **P0/P1**: **Immediately invoke** `ecw:risk-classifier --phase3` to execute Phase 3 calibration. Do NOT output "analysis complete, shall I calibrate?" or any confirmation. Mark the biz-impact-analysis Task as complete; if a pending "Phase 3 Calibration" Task exists, mark it `in_progress`.
> - **P2**: Suggest executing Phase 3 (not mandatory; user decides).
> - **P3**: No Phase 3 needed.
> - If `Auto-Continue` field is missing or `no` in session-state.md, fall back to waiting for user confirmation (backward compatibility).

If TaskList has a pending "Phase 3 Calibration" Task, marking biz-impact-analysis Task as completed will automatically unblock that Task.

## Subagent Ledger Update

After Agent returns, append one row to `.claude/ecw/session-data/{workflow-id}/session-state.md` Subagent Ledger table:

```
| biz-impact-analysis | analyst | ecw:biz-impact-analysis | opus | large | {HH:mm} | {duration} |
```

Note time before dispatch and compute duration after Agent return.

Scale reference: small (<20K tokens), medium (20-80K), large (>80K). biz-impact-analysis agent is typically large (needs to read multiple knowledge files + code scanning).

**Timeout**: 300s (agent reads multiple knowledge files and scans code). If Agent has not returned, terminate and offer retry (see Error Handling).

## Error Handling

| Scenario | Handling |
|----------|---------|
| Biz-impact-analysis Agent returns empty or fails | Record `FAILED` in Subagent Ledger → retry once → still fails: notify user `[DEGRADED: business impact analysis unavailable]` and suggest manual assessment |
| `ecw-path-mappings.md` missing | Agent cannot map files to domains → output `[Warning: path mappings not found, domain identification degraded]` and proceed with file-path-based heuristic grouping |
| Knowledge files missing (cross-domain-calls.md, mq-topology.md, etc.) | Agent logs `[Warning: {file} not found]` per missing file → analysis continues with available data, "Analysis Coverage" section in report reflects gaps |
| `git diff` returns empty | No changes to analyze → notify user and exit without dispatching agent |

## Common Rationalizations

| Your Thought | Reality |
|-------------|---------|
| "Changes are in one domain, no cross-domain impact" | Single-domain changes can affect shared resources consumed by other domains. Always check shared resources and end-to-end paths. |
| "The knowledge files are probably up to date" | Knowledge docs drift from code. Reverse validation catches stale entries. Skip it and the report includes phantom dependencies. |
| "No MQ or external references, so skip those sections" | Correct to skip if scan confirms no hits. But note it explicitly in Analysis Coverage — silent skips are the most dangerous blind spots. |
| "This is a low-risk query change, impact is minimal" | Query logic changes can affect downstream filtering. Medium risk, not minimal. Check end-to-end paths. |
| "Report is getting long but I need to include everything" | Conciseness is a hard constraint. Prioritize action items (max 3). Move verbose details to structured tables. |

## Notes

- Analysis results depend on dependency graph data quality under ecw.yml `paths.knowledge_common`
- The "Analysis Coverage" section in the report indicates which dimensions may have gaps
- "Unregistered cross-domain calls" flagged in the report are auto-backfilled to `cross-domain-calls.md` (conservative: append-only, no modification/deletion)
- "Suspected stale entries" flagged in the report still need manual confirmation before cleanup