outreach-enrich-company
Adds fresh detail to a thin saved company profile and drafts messages to its people. Use when the user wants it filled out or messages written ("enrich a card", "draft messages from a card"). Not for
SKILL.mdplugins/mz-biz-outreach/skills/outreach-enrich-company/SKILL.md
GitHubOutreach Enrich Company
Overview
You are an orchestrator that takes a company card produced by /outreach-research and does two things, in this order:
- Deepen the dossier. Fan out five specialist enrichment subagents in parallel —
outreach-news-finder,outreach-tech-analyst,outreach-growth-analyst,outreach-scanner,outreach-contact-finder— each instructed to find data not already in the card (gap-filling, not duplicating). Their findings are merged into a new## Deeper Intelligencesection appended to the card. - Draft outreach letters. Using the original card data plus the deeper intelligence as personalization fuel, draft one letter per named Key Contact via
expert-copywriter— each subtly calibrated to the contact's inferred communication style (Social Styles, produced grounded-only byoutreach-personality-profiler) — then mandatorily naturalize each viaexpert-naturalizer. The letters are appended as a## Outreach Letterssection inside the card.
The whole pipeline is fully automatic — no approval gates, no mid-run prompts. The user invokes the skill, the skill enriches and writes letters, the card now holds them both.
When to Use
Works on an existing company card (produced by /outreach-research, lives at .mz/outreach/<run>/companies/<slug>.md) to add richer company intelligence, outreach letters for its key contacts, or both.
When NOT to use
- The user has only a company name or domain and no card — run
/outreach-researchfirst (single-company mode if they want only one). - The user wants to discover companies first — use
/outreach-research. - The user only wants to record an action already taken on a card (an email sent, a call made) — use
/outreach-update-card. - The user wants recruiter contacts for a job application — use
/job-recruiter-infoin mz-job-outreach. - The user wants general marketing copy (landing pages, announcements, social) — use
/copywritefrom mz-creative. - The user wants letters written to a separate location than the card — this skill only writes back into the card.
Input
$ARGUMENTS— Required: an absolute or repo-relative path to a company card.mdfile. Two accepted shapes:- Research baseline:
.mz/outreach/<run>/companies/<slug>.md— direct output of/outreach-research. First activation moves it toactive/. - Active:
.mz/outreach/active/<YYYY-MM-DD>_<slug>.md— a card already activated by this skill or/outreach-update-card. The date prefix is the last interaction date; this run updates it to today.
- Research baseline:
- Optional modifiers (case-insensitive, order-independent):
channels:email|linkedin|both— overrides per-contact channel auto-detection. Default: best channel per contact (email if known, else LinkedIn DM, else skip).sender:<inline text or path>— overrides the sender voice. Inline text becomes the voice description; a path is read as a sender bio markdown file. Default: read<run>/strategy.jsonsender_voicefield if present.enrich:only— run the deep-enrichment phase and skip letter drafting (card gets## Deeper Intelligencebut no## Outreach Letters).enrich:skip— skip the deep-enrichment phase and draft letters directly from the existing card content only.brief:<run_name>— pull the local business-climate dossier from anoutreach-briefrun; the combo matching the card's sector and location seeds the letters with regional-climate angles. Default: unset (company data only).personality:off— disable reader-style calibration entirely (skip the Phase 3 profiling wave; letters use neutral structure). Default: on — role-based priorities always shape the leading angle, and footprint profiling runs automatically when a contact has a usable public profile.- Default (neither flag): both phases run.
If $ARGUMENTS is empty or no source can be resolved (see Source Resolution below), emit STATUS: BLOCKED with a clear message and stop.
Argument Parsing
The $ARGUMENTS variables and the Source resolution rules (which derive company_slug and source_path) are in ${CLAUDE_SKILL_DIR}/references/argument-parsing.md. Read it at Phase 0.
Directory Structure
The run's file layout (state, card snapshot, enrichment, briefs, profiles, drafts, card lifecycle) and the task_name and <run_dir> rules are in ${CLAUDE_SKILL_DIR}/references/directory-structure.md. Read it at Phase 0.
Constants
- PROJECT_ROOT: resolved once in setup with
echo "${MZ_MEMORY_ROOT:-$(git rev-parse --show-toplevel 2>/dev/null || pwd)}"; expand<PROJECT_ROOT>to it in every command and agent prompt. - SKILL_DIR:
${CLAUDE_SKILL_DIR}. Phase files write<SKILL_DIR>/...because files opened with Read get no variable substitution; expand it to this absolute path before reading a file, running a command, or passing a path to an agent.
Core Process
Phase Overview
| # | Phase | Agent(s) | Details |
|---|---|---|---|
| 0 | Setup + card validation | — (orchestrator) | phases/parse_card.md |
| 1 | Parse card + light web enrichment | — (orchestrator + WebSearch) | phases/parse_card.md |
| 2 | Deep enrichment via subagents | outreach-news-finder, outreach-tech-analyst, outreach-growth-analyst, outreach-scanner, outreach-contact-finder |
phases/deep_enrichment.md |
| 3 | Reader profiling + draft letters | outreach-personality-profiler, expert-copywriter |
phases/draft_letters.md |
| 4 | Naturalize letters | expert-naturalizer |
phases/naturalize_and_finalize.md |
| 5 | Rewrite card (deeper + letters) | — (orchestrator, Read + Write) | phases/naturalize_and_finalize.md |
| 6 | Verification summary | — (orchestrator) | verification-summary.md (reference) |
Read phases/parse_card.md when you reach Phase 0, phases/deep_enrichment.md when you reach Phase 2, phases/draft_letters.md when you reach Phase 3, and phases/naturalize_and_finalize.md when you reach Phase 4 (or Phase 5 when enrich_mode == "only"). If enrich_mode == "only", Phases 3 and 4 are skipped; Phase 5 still runs to append the ## Deeper Intelligence section. If enrich_mode == "skip", Phase 2 is skipped; Phase 5 only appends the ## Outreach Letters section.
Phase 0: Setup + card validation
Run from phases/parse_card.md; read it at the start of Phase 0.
Phase 1: Parse card + light web enrichment
Run from phases/parse_card.md, which Phase 0 already loaded.
Phase 6: Verification summary
After Phase 5 completes, emit the visible block defined in ${CLAUDE_SKILL_DIR}/references/verification-summary.md, which also lists when STATUS is DONE_WITH_CONCERNS.
State Management
State persists to .mz/task/<task_name>/state.md under the project root, ${MZ_MEMORY_ROOT:-$(git rev-parse --show-toplevel 2>/dev/null || pwd)}, resolved once at the start of the run. Schema is v2: line 1 is schema_version: 2; line 2 is Status: with one of pending, running, blocked, complete, aborted_by_user or failed, lowercase and unquoted; then Phase, Started, phase_complete (boolean) and what_remains (YAML list of strings), with skill-specific keys below them. Milestones go in Phase; the reason for blocked or failed goes in what_remains. Set phase_complete: false on phase entry and true once the phase's artifacts are written and its gates pass. Refresh what_remains on every phase transition; it is [] when Status is complete. Write the file after every phase and before every gate, and re-read it after compaction. Record decisions worth recalling under a heading that reads exactly ## Decisions, placed before ## Notes; mz-memory harvests them only once Status is complete. Under a heading that reads exactly ## Lessons, add one line whenever a review finding is accepted and fixed, a step fails, gate feedback corrects a mistake, or an approved approach is reversed, in the form - (review|failure|correction|change) WHEN <situation> → <what to do> — WHY <cause>, concrete and under about 40 words; mz-memory harvests lessons whatever the Status. Treat a legacy completed as complete. On reading a schema_version: 1 or unversioned file, put schema_version: 2 on line 1 and Status on line 2, add the missing keys, and log the upgrade. Leave a file with a higher schema_version untouched: ask whether to Restart (archive it), start New (_v2) or Abort, and stop if nobody can answer.
Red Flags
- You drafted a letter referencing a personalization hook that came from your own background knowledge rather than the card, the light-pass enrichment, or the Phase 2 deeper intelligence.
- You wrote a letter that names a product, customer, metric, or date not in the contact's
verified_entitieslist (which now includes Phase 2 findings). - You asserted a Social Style for a contact without a citable public-footprint cue. With no evidence the profiler must fall back to role-based priorities — never guess a style from the title alone.
- A letter named or alluded to the recipient's inferred personality or communication style. The reader calibration must stay invisible in the text.
- You skipped the naturalize pass for any draft. The naturalize pass is mandatory.
- You wrote the letters or the deeper intelligence to a separate file outside the card. Both outputs must be appended inside the card.
- You used a copywriter, naturalizer, or enrichment artifact without first checking that it exists on disk.
- You exceeded 6 concurrent agent dispatches in a single wave (Phase 2 uses 5 subagents; that fits in one wave).
- You dispatched an enrichment subagent without first reading its existing-card-data block from
card_parsed.json. Subagents must be told what is ALREADY known so they can hunt for gaps, not duplicate findings. - You added a
## Outreach Lettersor## Deeper Intelligencesection to the card without first removing the prior version (re-runs must replace, not duplicate). - You modified the card's earlier sections (Overview, Key Contacts, Outreach Recommendation, etc.) — only the new
## Deeper Intelligence+## Outreach Letterssections and their trailing footer lines are this skill's outputs. - You wrote the rebuilt card back to
source_pathinstead oftarget_path = <PROJECT_ROOT>/.mz/outreach/active/<today>_<company_slug>.md. The card must end this run at the active path with today's date prefix. - You preserved the source file at its research-baseline or stale-active location after Phase 5 completed (a successful run leaves exactly one card for the slug, at
target_path). - You left multiple active files for the same slug behind — only today's-dated file should remain.
- You discarded the
## Interaction Historysection (if present below the prior## Outreach Letterssection). Interaction history is owned by/outreach-update-cardand must be preserved verbatim.
Verification
Before emitting STATUS: DONE, confirm:
- The card now lives at
target_path = <PROJECT_ROOT>/.mz/outreach/active/<today>_<company_slug>.md. - If
source_path != target_path, the source file no longer exists. <PROJECT_ROOT>/.mz/outreach/active/contains exactly one file matching*_<company_slug>.md.- If
enrich_mode != "skip", the card attarget_pathcontains a## Deeper Intelligencesection with at least one populated subsection. - If
enrich_mode != "only", the card attarget_pathcontains a## Outreach Letterssection. Every non-skipped (contact, channel) pair produced one letter under that section. Each email letter shows a**Subject**:line; every letter opens with a salutation and closes with a voice-adaptive gratitude sign-off (Thanks,/Best regards,/Sincerely,) plus the sender's first name. No letter body contains em-dashes, en-dashes, curly quotes, or the ellipsis character. Whenpersonality_mode != "off", each letter carries a*Reader calibration: ...*provenance line and no letter names or alludes to the inferred trait. The## Naturalization reportblock does not appear under any letter in the final card. - The card retains its original
*Generated: <date> | Sources: <list>*footer above the new sections. - New footers sit at the bottom of each new section:
*Deeper intelligence: <date> | subagents: <N> | failed: <M>*and*Letters generated: <date> | naturalized: yes | light queries: <K> | reverted: <R>*. - Section order on the card is: original sections →
## Deeper Intelligence(if present) →## Outreach Letters(if present) →## Interaction History(if it existed before).
References
${CLAUDE_SKILL_DIR}/references/argument-parsing.md: the$ARGUMENTSvariables and the Source resolution rules that turnuser_pathintocompany_slugandsource_path. Read it in full at Phase 0.${CLAUDE_SKILL_DIR}/references/directory-structure.md: where state, the card snapshot, enrichment artifacts, briefs, profiles and drafts live, the card lifecycle, and thetask_nameand<run_dir>rules. Read it in full at Phase 0.${CLAUDE_SKILL_DIR}/references/verification-summary.md: the final verification block and theSTATUSrule. Read it in full at Phase 6.
Resume Support
Before creating anything in Phase 0, run the resume check in phases/parse_card.md (section Resume Support); it maps the Phase value in state.md to the phase to resume from.
Error Handling
- User path missing or no source can be resolved (user path does not exist AND no matching active card found) →
STATUS: BLOCKED. Surface the failing check. strategy.jsonmissing → fall back to default sender voice; final STATUS becomesDONE_WITH_CONCERNS; the verification block names the absence.- Light-pass cap hit (6 calls) → drop the lowest-priority queries; record the drop in
state.mdand the final footer. - One enrichment subagent in Phase 2 returns
BLOCKEDorNEEDS_CONTEXT→ continue the wave for other subagents; mark the corresponding subsection asunavailablein the## Deeper Intelligenceblock; final STATUS becomesDONE_WITH_CONCERNS. Do not auto-retry. - All five enrichment subagents fail → still emit the
## Deeper Intelligencesection with each subsection markedunavailable: <reason>so the failure is visible in the card; downgrade STATUS toDONE_WITH_CONCERNS. - One
expert-copywriterdispatch returnsBLOCKEDorNEEDS_CONTEXT→ continue the wave for other contacts; mark this contactfailed_draftinstate.md; the contact appears as### Skipped: <Name> — <reason>under the card section; final STATUS becomesDONE_WITH_CONCERNS. Do not auto-retry — the failure reason must surface to the user. - One
expert-naturalizerdispatch over-shortens a draft (post-validation finds word count outside channel budget by >5% or subject line drifted) → restore the pre-naturalize snapshot; marknaturalize_revertedinstate.md; the pre-naturalize draft is what gets appended; final STATUS becomesDONE_WITH_CONCERNS. - Card has no Key Contacts at all AND
enrich_mode != "only"→STATUS: BLOCKEDwith message "card has no Key Contacts — re-run /outreach-research or hand-edit the card first, or run with enrich:only to skip letters." - Any agent run more than once with
NEEDS_CONTEXT→ escalate viaAskUserQuestionwith the agent's stated missing piece. Otherwise no user prompts. - Source-file deletion fails during Phase 5 cleanup (permission, file vanished, fs error) → do not fail the run; log
relocation_cleanup_failed: <reason>instate.md; downgrade final STATUS toDONE_WITH_CONCERNS. The card content attarget_pathis canonical regardless; the user can clean up the stale source manually. .mz/outreach/active/is unwritable (permission, read-only mount) →STATUS: BLOCKED. The skill cannot complete without writing the activated card.- Never fabricate contacts, hooks, news items, metrics, or any other content. Incomplete output is better than fabricated content.