| name | client-sourced-deal-review-memo |
| description | Family office: Produce a review memo for a deal the principal sourced themselves. Captures deal sponsor, structure, terms, conflicts, and recommendation. |
Family Office · Client-Sourced Deal Review Memo
For Claude: How to Use This Skill
Skill instructions are preloaded in context when this skill is active. Do not
perform filesystem searches or tool-driven exploration to rediscover them; use
the guidance below directly.
When to Use
Invoke when the advisor asks about:
- client sourced deal
- principal brought a deal
- friend referral investment
- review this deal
Artifact Specification
This skill produces a single named deliverable:
- Artifact:
Client-Sourced Deal Review Memo
- Format at this iteration: markdown (
artifact.md) plus a structured
interview.json capturing every advisor input. PDF / DOCX / XLSX companion
renders and DMS / Snowflake push land in the execution-pipeline PR tracked
under issue #427.
The artifact is written to a skill-local path, not a global directory:
<invocation-cwd>/artifacts/family-office/client-sourced-deal-review-memo/<YYYYMMDD-HHMMSS>/
artifact.md
interview.json
manifest.json
Interview Inputs
Minimum-viable interview — the skill asks only what is needed to personalize
the deliverable. Pre-fill rules against family memory land in the execution-
pipeline PR.
deal_name — Deal name?
sponsor — Deal sponsor?
investment_amount — Proposed investment amount?
structure — Deal structure (equity / debt / convertible)?
key_terms — Key terms (pref / cap / board / vesting)?
conflicts — Any conflicts of interest?
recommendation_lean — Lean recommendation (proceed / decline / pause)?
Workflow
- Run
python scripts/agent.py --config config.json (or invoke via Claude
Code with a config blob).
- The agent validates config, runs the interview (TTY or fixture-driven),
and produces the artifact under the canonical local path.
- The agent writes
manifest.json with artifact metadata (name, version,
content hash, pillar, skill_name, created_at).
- The agent optionally writes memory entries to the knowledge skill's
memory_objects table via psycopg if config.memory_dsn is provided.
Without a DSN, memory writes are skipped cleanly.
Memory Conventions
Memories written by this skill are tagged with:
subject = the artifact name
source = "client-sourced-deal-review-memo"
memory_type ∈ {decision, assumption, commitment, open_question}
Security & Confidentiality
- Never log interview answers or artifact contents at INFO level.
- Never include SSN, EIN, account numbers, or full financial amounts in
log lines. If a WHERE-clause field carries such data, log only a sha256
hash.
- The artifact directory is local-only at this iteration. DMS push (with
confidentiality-label routing) is handled by the execution-pipeline PR.
Reference