| name | Cold Email Taste Review |
| description | Child skill for scoring whether a cold email draft is specific, proportionate, credible, and reputation-safe enough for a serious sender to send — including "tighten this up" or "is this good?" requests on an existing draft. 8-dimension scorecard, auto-fails, screenshot test, and named rewrite moves per failure; routes structural failures to the compiler instead of line-editing. |
| skill_type | strategy |
| altitude | domain |
| activation | progressive |
| parent_id | cold_email_engagement_first_outreach |
| depth | 1 |
| preserve_markdown | true |
| legacy_paths | ["cold_email_outreach.taste_review","cold_email_outreach.shame_function"] |
| reference_modules | [{"id":"cold_email_taste_review.fake_warmth_and_rewrites","name":"Fake-Warmth Detector and Rewrite Patterns","summary":"Named fake-warmth patterns, subject/preview rejection rules, five bad-to-good rewrite pairs, and expanded mode-dignity checks for PR, recruiting, research, and investor modes.","when_to_load":["When the draft fails dimensions 1, 2, 7, or 8, or any auto-fail fires.","When the user asks why an email is bad or how to fix a failed dimension."],"path":"references/fake-warmth-detector-and-rewrites.md","visibility":"internal"}] |
| path | apps/web/src/lib/services/agentic-chat/tools/skills/definitions/cold_email_taste_review/SKILL.md |
Cold Email Taste Review
Identity
Use this child skill when reputation risk is the main question. The north star is qualified conversations started per unit of market trust consumed — never emails sent. A score without a named fix is not a review.
This is a strategy skill at domain altitude: its job is to decide — grade a draft against a sourced rubric and return a verdict with a named fix — not to compose or rebuild the email. Structural rebuilds are handed to sibling skills (see Routing).
Activation
- The draft may sound generic, pushy, automated, or fake-warm
- The personalization may be decorative
- The ask may exceed the trust earned
- Proof, claims, or tone may embarrass the sender if screenshotted
- The user asks whether the email is good, tasteful, or safe to send
Judgment
The decision spine. Score all eight dimensions, then check auto-fails, then run the screenshot test, then map to a verdict. A score without a named fix is not a review.
Taste Scorecard
When grading any draft: score all eight dimensions, then check auto-fails, then run the screenshot test, then map to a verdict. A score without a named fix is not a review.
Framing metric for every verdict (keep verbatim):
qualified conversations started per unit of market trust consumed
Never grade an email on whether it will get sent or opened at volume. Grade it on whether it starts a qualified conversation without burning trust the sender cannot recover.
The Scorecard (8 dimensions, 0–2 each, 16 max)
Score 2 = pass, 1 = partial, 0 = fail. Every 0 or 1 must be reported with the failed dimension named and a specific rewrite move attached.
| # | Dimension | 2 (pass) | 0 (fail) | Source |
|---|
| 1 | Two-people test (personalization honesty) | Anchor line could only be about this recipient, OR is honestly graded as 1:many relevance with a strong trigger-based signal | "Personalized" opener that could be sent to two people unchanged; decorative detail | Becc Holland: "If you can send the same email to two people, it is relevance — not personalization" |
| 2 | Bridge integrity | Remove the hook and the outreach reason collapses — the anchor causes the email | Hook is removable; personal detail with no business consequence ("fellow eagle," shared school alone) |
