claude-skills/productivity/email/skills/inbox-setup/scripts
Claude a2e9e48eb2
feat(productivity): email pair (inbox-setup + inbox-triage) — Path-B workflow-pair slice
Slice 3 of 13: workflow-pair shape. Validates the Path-B conversion pattern
for two coupled skills sharing a strict 7-file KB contract. Also introduces
the productivity/ domain folder per the navigation-map distinction in
CLAUDE.md (engineering/ = software-engineering scope; productivity/ =
generic productivity workflows).

DOMAIN FOLDER DECISION

CLAUDE.md defines engineering/ as "Engineering (POWERFUL) — Agent design,
RAG, MCP, CI/CD, database, observability." Email triage is generic
productivity, not software engineering. New folder: productivity/.

Capture (Slice 1, merged in PR #659) was placed under engineering/ before
this distinction was sharpened. It will move to productivity/ in a separate
cleanup PR — moving a merged plugin in this slice would risk breaking
anyone who installed it from engineering/capture/.

Future productivity slices (02-reflect) will go under productivity/ from
the start.

SOURCE SPECS

- megaprompts/06-inbox-setup-megaprompt.md (PR #657)
- megaprompts/07-inbox-triage-megaprompt.md (PR #657)

The megaprompts are canonical; these plugins are working implementations.
PR #657's cross-skill consistency audit verified the 7 KB filenames
align verbatim between the two megaprompts. This slice preserves that
alignment.

WHAT THE PAIR DOES

Two coupled skills sharing a 7-file KB at ${WORKSPACE}/Email/:

  inbox-setup (run once):
    Interactive 8-section interview (~25-31 grill-me questions) →
    writes 7 KB files (taxonomy, patterns, evaluation-framework,
    rate-card, blocklist, tracker, triage-log/).

  inbox-triage (run recurringly):
    Light-intake (max 2 optional override questions). Reads 7 KB files,
    classifies recent emails, researches new senders, generates
    recommendations (TAKE IT / WORTH / PASS / FLAG), drafts replies
    (NEVER SENDS), delivers report, updates blocklist + tracker, writes
    per-run log. 10 execution steps.

PATH-B CONVERSION DISCIPLINE

- Both megaprompts' frontmatter descriptions preserved verbatim.
- Both workflow structures preserved 1:1 in respective SKILL.md files.
- All 8 setup sections preserved verbatim with per-question structure
  (S{n}.Q{m}) + "why I'm asking" rationale.
- All 10 triage steps preserved verbatim.
- DRAFTS-ONLY rule preserved + amplified (stated in SKILL.md, agent,
  command, AND enforced by draft_safety_validator.py).
- Skip-logic preserved (S4 conditional on S1 surfacing opportunities).
- 7-file KB contract referenced verbatim in both directions.

REPO STRUCTURE — MULTI-SKILL LAYOUT (CLAUDE.md plugin-schema rule)

productivity/email/
├── .claude-plugin/plugin.json    ← skills: ["./skills/inbox-setup", "./skills/inbox-triage"]
├── README.md                      ← pair overview + 7-file contract diagram
├── agents/
│   ├── cs-inbox-setup.md          ← interview persona
│   └── cs-inbox-triage.md         ← recurring-run persona, DRAFTS-ONLY enforcer
├── commands/
│   ├── cs-inbox-setup.md
│   └── cs-inbox-triage.md
└── skills/
    ├── inbox-setup/
    │   ├── SKILL.md               ← 8 sections, 25-31 Q discipline
    │   ├── references/
    │   │   ├── kb_file_contract.md            ← write-side spec
    │   │   ├── grill_me_section_walk.md       ← discipline + skip-logic
    │   │   └── voice_calibration.md           ← sample-extraction theory + 7 sources
    │   └── scripts/
    │       ├── kb_validator.py                 ← stdlib: 7-file contract check
    │       ├── section_progress_tracker.py    ← stdlib: 8-section walk state
    │       └── voice_sample_analyzer.py        ← stdlib: pattern extraction
    └── inbox-triage/
        ├── SKILL.md               ← 10 steps + DRAFTS-ONLY rule
        ├── references/
        │   ├── kb_file_contract.md            ← read-side spec (mirror)
        │   ├── triage_decision_framework.md  ← TAKE/WORTH/PASS/FLAG + 7 sources
        │   └── drafts_only_safety.md         ← NEVER-SEND canon + 7 sources
        └── scripts/
            ├── kb_reader.py                    ← stdlib: parsed KB load + fail-fast
            ├── search_window_calculator.py    ← stdlib: cadence → window
            └── draft_safety_validator.py      ← stdlib: post-run NEVER-SEND check

20 files, 3,710 lines. Roughly 2x a single-skill slice (capture: 1,560,
pulse: 1,643), appropriate for two coupled skills.

VERIFIED CLEAN

Smoke tests on all 6 scripts:

- kb_validator.py: 15/15 PASS on sample (4 core files + h1s + sections +
  conditional file expectations + triage-log/ dir)
- section_progress_tracker.py: full lifecycle (start → record_q → record_
  section_done → record_skip → status). Active section advances correctly
  past S4 skip.
- voice_sample_analyzer.py: 5 samples → register/length/hedging/I-vs-We
  verdicts + opening + sign-off pattern extraction + email-patterns.md
  output block generation.
- kb_reader.py: reads 5/6 sample files (rate-card.md correctly absent),
  PASS verdict, structured parsing.
- search_window_calculator.py: 2x-daily + 14:00 → 9h lookback,
  window_start 05:00, run_label "Afternoon". Provides Gmail/Outlook/IMAP
  query templates.
- draft_safety_validator.py: PASS on clean log; FAIL on log with
  `gmail.users.messages.send` (caught by 2 patterns — defense in depth).
  Action-required guidance fires.

CROSS-SKILL CONTRACT ALIGNMENT (PR #657 audit verbatim alignment preserved)

Each of the 7 KB filenames referenced multiple times on both sides:
  email-taxonomy.md:      setup=5 triage=9
  email-patterns.md:      setup=4 triage=8
  evaluation-framework:   setup=5 triage=7
  rate-card.md:           setup=5 triage=4
  blocklist.md:           setup=4 triage=6
  tracker.md:             setup=4 triage=7
  triage-log:             setup=6 triage=7

VERTICAL-SLICE STATUS

  ✓ Slice 1: capture (light prompt-flow, PR #659 merged)
  ✓ Slice 2: pulse (research-pack, PR #660 merged)
  ✓ Slice 3: email-pair (workflow-pair — this PR; introduces productivity/)
  ☐ Slice 4: generator (04-landing) — validates Next.js code template emission
  ☐ Slice 5: orchestrator/router (13-research) — must reconcile with existing
    engineering/autoresearch-agent/

After Slice 4 validates the generator shape, only the orchestrator shape
remains to be validated. The 6 remaining research-pack skills (litreview,
grants, syllabus, patent, dossier, notebooklm) and 02-reflect can then be
batched in a single PR each.

NOT DONE IN THIS PR (intentional)

- .claude-plugin/marketplace.json not updated (separate concern; done after
  all 13 ship)
- .codex/skills/inbox-setup + .codex/skills/inbox-triage symlinks not added
  (auto-sync workflow handles on merge)
- engineering/capture/ NOT moved to productivity/capture/ (would break
  anyone who installed from current path; address in separate cleanup PR)

https://claude.ai/code/session_01FEUmeuYhmnxVFq7EZM8ZSw
2026-05-15 15:28:00 +00:00
..
kb_validator.py feat(productivity): email pair (inbox-setup + inbox-triage) — Path-B workflow-pair slice 2026-05-15 15:28:00 +00:00
section_progress_tracker.py feat(productivity): email pair (inbox-setup + inbox-triage) — Path-B workflow-pair slice 2026-05-15 15:28:00 +00:00
voice_sample_analyzer.py feat(productivity): email pair (inbox-setup + inbox-triage) — Path-B workflow-pair slice 2026-05-15 15:28:00 +00:00