git:20261004.58fdc08 to git:20261004.0172f06

+ description: "Signal-anchored B2B cold email: first touch under 90 words with a binary ask, follow-ups, subject lines, spam lint, domain deliverability, send-list hygiene, reply triage, and outbound audits. Use when the user asks for a cold email, follow-up sequence, deliverability or spam check, reply scoring, or send-list cleanup. Not for give-first offers (sales-offer), picking who to contact (prospect-list), or post-opt-in email (email-sequence)."

+ argument-hint: "[write | sequence | subject | lint | deliverability | list | reply | nurture | audit | rhythm | status | setup]"

+ allowed-tools: Read Write Grep Glob WebFetch Bash(python3 ${CLAUDE_PLUGIN_ROOT}/scripts/score_letter.py:*) Bash(python3 ${CLAUDE_PLUGIN_ROOT}/scripts/spam_word_lint.py:*) Bash(python3 ${CLAUDE_PLUGIN_ROOT}/scripts/score_subject_line.py:*) Bash(python3 ${CLAUDE_PLUGIN_ROOT}/scripts/check_deliverability.py:*) Bash(bash ${CLAUDE_PLUGIN_ROOT}/scripts/dig_dns.sh:*) Bash(python3 ${CLAUDE_PLUGIN_ROOT}/scripts/score_list.py:*) Bash(python3 ${CLAUDE_PLUGIN_ROOT}/scripts/score_reply.py:*) Bash(dig:*) Bash(host:*) Bash(nslookup:*)

+

+ run the setup mode ([modes/setup.md](modes/setup.md)) before drafting

+ real outreach. If `operator` or `icp` is missing, say "Run `/gtm:setup`

+ once for the whole suite" first. Offer once; if they decline, say the

+ output will be framework-shaped but generic.

+ ## Modes — route by $ARGUMENTS

+ Run the preflight above first, every time. Then pick one mode:

+ - If `$ARGUMENTS` starts with a mode name below, go straight to that mode

+ (the rest of `$ARGUMENTS` is its input).

+ - If `$ARGUMENTS` is empty, run `status`: it reads the project and names

+ the one next step.

+ - Otherwise match the user's words to the table. If intent is still

+ ambiguous, ask one clarifying question — never two.

+

+ Read the mode file with the Read tool and follow it.

+

+ | You say / argument | Mode file |

+ |---|---|

+ | (nothing), `status`, "where do I start", "what's next" | [modes/status.md](modes/status.md) |

+ | `setup`, "set up my brand config", "configure my voice" | [modes/setup.md](modes/setup.md) |

+ | `write`, "write a cold email to...", "rewrite this cold email" | [modes/craft.md](modes/craft.md) |

+ | `sequence`, "build a follow-up sequence", "Day 3 / 7 / 14" | [modes/craft.md](modes/craft.md) (sequence mode) |

+ | `subject`, "subject lines for...", "critique this subject" | [modes/subject.md](modes/subject.md) |

+ | `lint`, "spam-check this", "will this land in spam" | [modes/lint.md](modes/lint.md) |

+ | `deliverability`, "is my domain ready", SPF / DKIM / DMARC | [modes/deliverability.md](modes/deliverability.md) |

+ | `list`, "score / clean my send list", "dedup this CSV" | [modes/list.md](modes/list.md) |

+ | `reply`, "score this reply", "triage my replies" | [modes/reply.md](modes/reply.md) |

+ | `nurture`, "re-engage cold or stalled prospects" (not opted in) | [modes/nurture.md](modes/nurture.md) |

+ | `audit`, "audit my outbound program" | [modes/audit.md](modes/audit.md) |

+ | `rhythm`, "this week's queue", "Friday review" | [modes/rhythm.md](modes/rhythm.md) |

+ | "Review / score this draft" | `cold-email-reviewer` agent |

+

+ Files the modes write in the operator's project go under `gtm/` at the

+ project root (create it if missing): `gtm/letter.json` (the draft:

+ `public_signal`, `subject`, `letter`), `gtm/send-list.csv` (the send

+ list; `--write` adds `.cleaned.csv` and `.removed.csv` beside it), and

+ `gtm/replies.jsonl` (one reply per line). `brand-config.json` and

+ `SOUL.md` stay at the project root.

+

+ ## Variables (every mode uses these)

+ reuse across every mode:

+ Write the draft to `gtm/letter.json`, then:

+

+ - `score_letter.py --file gtm/letter.json`: exit 0 (signal quoted, no

+ - `spam_word_lint.py --file gtm/letter.json`: score 75 or higher

+ - `score_subject_line.py --file gtm/letter.json`: score 70 or higher

+ On exit 1 each reason reads `- what is wrong → what to change`. Apply

+ every change and run it again; do not show a draft that fails.

+

+ The scripts live in `scripts/` at the plugin root. Call them as

+ `${CLAUDE_PLUGIN_ROOT}/scripts/<name>.py`, which resolves to the

+ installed plugin from any project directory. Outside Claude Code,

+ `python3 scripts/<name>.py` from the repo root is the same script. All

+ are Python 3.8+ stdlib only; DNS checks need `dig`.

+ Every scorer takes `--file PATH` or `--stdin`, prints text by default and

+ one JSON object with `--json`, and exits 0 pass, 1 refused (each reason

+ with its fix, then `Next: fix the lines above and run this again.`), 2

+ bad input. On a pass the last line names the next step.

+

+ | Script | Mode |

+ | `score_letter.py` | craft, self-check (before the spam lint) |

+ | `spam_word_lint.py` | lint, self-check |

+ | `score_subject_line.py` | subject, self-check |

+ | `score_reply.py` | reply |

+ | `score_list.py` | list |

+ | `check_deliverability.py`, `dig_dns.sh` | deliverability |

+ - **After this:** email-sequence (`/email-sequence:lifecycle-email`), when a prospect replies and opts in. End a successful run with that `Next:` line.

+ - [references/audit-rubric.md](references/audit-rubric.md) — The 30-point outbound audit rubric

+ - [SOUL.md](../../SOUL.md) — Operator voice template (the shape the setup mode fills in)