| name | knowledgebase-kickoff |
| description | Bootstrap a project knowledge base — folder skeleton, entity-page design, SCHEMA/CLAUDE/README scaffolding, an ingest-sources skill, .gitignore, git init, and initial commit. Use when the user wants to set up a knowledge base or queryable source of truth for a stakeholder-heavy project with meetings, emails, and documents. |
Kickoff Knowledge Base Project
Bootstrap a new project following the knowledge-base-per-project pattern: raw sources → LLM-maintained entity pages → queryable, auditable source of truth.
When to use
Use this skill when the user wants to set up knowledge management for a project that:
- Involves many stakeholders, frequent meetings, and lots of emails/documents.
- Requires supplier or third-party accountability tracking.
- Needs a queryable source of truth maintained incrementally by an AI agent.
Pre-flight
-
Determine target directory. Default: current working directory. If it's non-empty OR already contains any of the scaffolding files (SCHEMA.md, CLAUDE.md, AGENTS.md, README.md), STOP and confirm with the user before proceeding.
-
Gather project values. Ask the user for (or derive from conversation context):
PROJECT_NAME — human-readable, e.g., "Warehouse Automation Rollout"
PROJECT_DESCRIPTION_SHORT — one sentence, ≤ 30 words. What the project is.
PROJECT_DESCRIPTION_LONG — a paragraph (80–150 words) covering scale, stakeholders, timeline, and what makes this project complex.
OWNER_NAME — project owner's full name.
OWNER_EMAIL — project owner's email.
TEAM_DETAILS — a paragraph describing who's on the user's side, who the delivery partner is, and any other relevant structure.
Confirm the values with the user before writing files.
What to create
Mirror this skill's templates/ tree (resolved relative to this SKILL.md) into the target directory, substituting placeholders:
Folders
sources/{emails,transcripts,documents,ingested/{emails,transcripts,documents}}
people/
organizations/
workstreams/
meetings/
.claude/skills/ingest-sources/
Files (from templates/)
README.md
CLAUDE.md — copy to AGENTS.md as well (identical)
SCHEMA.md
LOG.md
PLAN.md — the PM action checklist
SOURCES-CHECKLIST.md — the data-collection audit
.gitignore
.claude/skills/ingest-sources/SKILL.md
Placeholders
Template files contain {{PLACEHOLDER}} tokens. Substitute with user-provided values before writing to the target:
| Token | Value source |
|---|
{{PROJECT_NAME}} | User-provided project name |
{{PROJECT_DESCRIPTION_SHORT}} | User-provided one-line description |
{{PROJECT_DESCRIPTION_LONG}} | User-provided context paragraph |
{{OWNER_NAME}} | User-provided owner name |
{{OWNER_EMAIL}} | User-provided owner email |
{{TEAM_DETAILS}} | User-provided team paragraph |
Post-creation steps
git init -q in the target directory.
git add -A and commit with message chore: initial knowledge base skeleton (Co-Authored-By: Claude).
- Report to the user:
- What was created.
- Confirmed owner + project name.
- Suggested next action: drop the first sources into
sources/{emails,transcripts,documents}/ and invoke the ingest-sources skill.
Maintenance
The templates/ tree in this skill is the canonical scaffolding — there is no external reference project. When work in a bootstrapped knowledge base improves its scaffolding in a way that generalizes, fold the improvement back into the templates here, re-inserting {{PLACEHOLDERS}} where project-specific values were substituted. This is the same compounding loop as the build skill's Close phase.
What this skill does NOT do
- Does not create entity pages speculatively — people, workstreams, organizations emerge from data during ingestion.
- Does not ingest any sources — that is the
ingest-sources skill's job, invoked after kickoff.
- Does not overwrite existing scaffolding. If any target file already exists, STOP and confirm.
- Does not push to any remote.
Output to user after kickoff
Report concisely:
- Target directory.
- Files created (brief list).
- Git commit hash.
- Next action: drop sources into
sources/ and invoke ingest-sources.