locus-taxy/config-drift

Purpose

Detect divergences between a market's config and the global reference standard. Separate intentional localisations (documented) from configuration accidents (undocumented). Risk-score undocumented findings. Drive resolution through Jira tasks.

Two modes, one skill:

• SA mode — single market, full detail, SA can override risk tier before output is written
• Director mode — all markets, rollup digest, no per-market Slack noise

Also runs inline as Dimension 6 of /go-nogo — see go-nogo skill.

---

Key Concepts

Two Divergence Categories

| Category | Definition | Action | |----------|-----------|--------| | Documented | Intentional localisation recorded in market-config-overrides.md or market-decision-log.md | Shown in report, no Jira task | | Undocumented | Setting differs from global with no record of why | Risk-scored, Jira task created, shown in report |

Risk Tiers (Claude assigns, SA can override in interactive mode)

| Tier | Field types | Rationale | |------|-------------|-----------| | 🔴 High | Auth method, SLA thresholds, critical field mappings (SHIP_CONFIRM, order status) | Silent failures, compliance risk, or data integrity impact | | 🟡 Medium | Carrier dispatch timeout, API version, retry policy, lead times | Operational impact but catchable in monitoring | | 🟢 Low | Time window labels, cosmetic field differences, non-critical config flags | No operational impact, low urgency |

Mode Detection

/config-drift [market]   → SA Mode (one market, full detail)
/config-drift            → Director Mode (all markets, rollup)
/config-drift all        → Director Mode (explicit)

---

Facilitation Source of Truth

When running interactively, use workshop-facilitation as the interaction protocol for pacing and interruption handling.

This file defines the workflow and domain outputs. If there is a conflict, follow this file's logic.

Application

Use template.md for the findings table structure.

---

SA Mode — Single Market

Stage 1: Auto-Diff

Read the following in parallel:

| Source | What to read | |--------|-------------| | Solution/Global-Solution-Document.md | Global config baseline — routing, SLAs, promise logic, returns | | Solution/Client-Configuration.md | Global client config baseline — carrier settings, auth, field mappings | | Solution/Markets/[market]/market-config-overrides.md | Market config deviations declared | | Solution/Markets/[market]/market-client-config.md | Market client config deviations | | Solution/Markets/[market]/carrier-integration-spec.md | Market integration settings — protocol, auth, field mappings, API version | | Solution/Markets/[market]/market-decision-log.md | Recorded decisions and documented exceptions |

Stage 2: Classify & Risk-Score

For each field that differs between market and global:

1. Check market-config-overrides.md and market-decision-log.md — is this divergence explicitly recorded?

   • Yes → Category: Documented. No Jira task.
   • No → Category: Undocumented. Assign risk tier + one-line rationale.

2. For undocumented items, present findings to the SA before writing output:

I found [n] divergences in [Market] ([n] undocumented, [n] documented).

Here are my risk assessments for the undocumented items — override any tier before I write the output:

  Field: carrier_dispatch_timeout
  Global: 30s | Market: 45s
  Risk: 🟡 Medium — operational impact if carriers have strict SLA windows
  Override? (press Enter to accept, or type Low/High)

  Field: auth_method
  Global: OAuth2 | Market: API key
  Risk: 🔴 High — API key auth has no expiry enforcement; token rotation risk
  Override? ...

3. Confirm overrides, then proceed to Stage 3.

Autonomous weekly scan: Skip the override step — write output immediately with Claude-assigned tiers.

Stage 3: Output (simultaneous)

Repo file → outputs/config-drift/[market]-[YYYY-MM-DD].md

---
market: [IKEA-XX]
mode: single
generated: [YYYY-MM-DD]
sa_lead: [name]
total_divergences: [n]
undocumented: [n]
documented: [n]
high_risk: [n]
medium_risk: [n]
low_risk: [n]
jira_tasks_created: [n]
---

Body: findings table (see template.md).

Jira tasks — one per undocumented divergence:

• Title: [Config Drift] [Market] — [field] differs from global standard
• Body: field, global value, market value, risk tier, rationale, resolution options
• Assignee: market SA
• Link to: market go-live epic

Slack post — to market delivery channel:

🔍 CONFIG DRIFT — [Market] | [date]

[n] divergences found: [n] undocumented, [n] documented

UNDOCUMENTED (action required):
  🔴 [field] — [one-line rationale]  → Jira: [task-id]
  🟡 [field] — [one-line rationale]  → Jira: [task-id]
  🟢 [field] — [one-line rationale]  → Jira: [task-id]

DOCUMENTED (intentional):
  ✅ [field] — [decision reference]

Full report: outputs/config-drift/[market]-[date].md

---

Director Mode — All Markets

Stage 1: Run SA-mode diff for each market (parallel)

Read all market folders under Solution/Markets/. Run the full diff for each. Do not present per-market override prompts — use Claude-assigned tiers directly.

Stage 2: Aggregate findings

Classify each market:

• 🟢 Clean — 0 undocumented divergences
• 🟡 Minor drift — 1–3 undocumented divergences
• 🔴 Significant drift — >3 undocumented divergences

Identify all High-risk undocumented items across all markets.

Stage 3: Output

Repo rollup → outputs/config-drift/rollup-[YYYY-MM-DD].md

YAML frontmatter with per-market summary + full findings for each market in body.

Slack digest — one post to S&D delivery channel only (no per-market posts):

📊 WEEKLY CONFIG DRIFT — all markets | [date]

🟢 Clean (0 undocumented):   [markets]
🟡 Minor drift (1–3):        [market] ([n]), [market] ([n])
🔴 Significant drift (>3):   [market] ([n])

Total undocumented divergences: [n] across [n] markets
🔴 High-risk items: [n]
  • [Market]: [field] — [rationale]
  • [Market]: [field] — [rationale]

@[Director] — full rollup: outputs/config-drift/rollup-[date].md
Jira tasks created: [n] (assigned to market SAs)

---

Common Pitfalls

Pitfall 1: Treating all drift as bad

Symptom: SA panics at any divergence; flags intentional localisations as blockers. Fix: Documented divergences are expected and healthy. The skill separates them. Only undocumented drift needs action.

Pitfall 2: Documenting drift retroactively to avoid a Jira task

Symptom: SA adds a vague entry to market-decision-log.md with no rationale just to make the finding go away. Fix: Documentation must include: what the deviation is, why it was approved, and who approved it. A one-word entry doesn't count.

Pitfall 3: Running config-drift only at go-live

Symptom: Drift accumulates across weeks; large number of findings at T-2 creates pressure to dismiss them. Fix: Run weekly (autonomous) to keep the surface small. The T-2 go-nogo check should be a confirmation, not a discovery.

Pitfall 4: Skipping the SA override step under time pressure

Symptom: SA approves all risk tiers without reviewing, then disputes Jira task priorities later. Fix: The override step is 2 minutes. If Claude assigned 🔴 High to a field the SA considers Low, fix it now — not after the Jira task is already assigned.

---

References

Connected skills

• skills/go-nogo/SKILL.md — /config-drift runs inline as Dimension 6 of the go-nogo gate
• skills/raid-scan/SKILL.md — Undocumented High-risk drift should be cross-referenced against open RAIDs

Repo sources

• Solution/Global-Solution-Document.md — global config baseline
• Solution/Client-Configuration.md — global client config baseline
• Solution/Markets/[market]/ — all market-level config and decision files
• outputs/config-drift/ — all past drift reports (machine-readable + human-readable)

Autonomous operations

Registered in CLAUDE.md under Autonomous Operations — runs weekly, no trigger needed. Output: repo rollup + Slack digest + Jira tasks.

---

Skill type: Interactive + Autonomous Placement: skills/config-drift/ SA mode output: outputs/config-drift/[market]-[YYYY-MM-DD].md + Jira tasks + Slack to market channel Director mode output: outputs/config-drift/rollup-[YYYY-MM-DD].md + Slack to S&D channel