← Back to insights

From chat histories to a single source of truth

Your team’s best prompts live in seventeen places. An individual chat history. A screenshot pasted into Slack. A Notion page someone made and stopped updating. The new starter rebuilds half the prompts from scratch, and the team’s craft does not compound.

The schema itself is the prompt template schema page; a sister post on why prompt libraries do not survive a hand-off covers what makes a single template runnable. This post is the layer up — what your team’s prompt library looks like as a whole.

The artefact stack

A prompt library that stops being tribal knowledge has five layers:

  1. A directory, not a wiki. A folder under version control with one file per template. Files are markdown or YAML; the schema is in the standards page. Notion is for prose; the prompts live in a repo.
  2. A worked example file per template. The example is the runnable test. Without it, “this template works” is a claim no one can verify; with it, a new team-mate substitutes their inputs and sees the shape of output to expect.
  3. A version tag on every template. Semantic versioning — v1.0, v1.1 — or a date suffix. Templates are artefacts, and every artefact your team produces has a version.
  4. A changelog. One line per change, dated. “Tightened H2 cap from ten words to seven words; example output regenerated.” The changelog is the audit trail when the brief that landed last quarter does not land this quarter.
  5. A review cadence. Every quarter, every template gets a self-review against the schema’s six fields (title, declared variables, body, worked example, expected output, failure modes). Templates that fail self-review are flagged or retired.

The five layers are independently cheap and structurally compounding. A team that ships all five at once does not need to revisit any of them for a year.

Why this is the same fix you have already shipped

Every artefact your team produces — briefs, decks, emails-in-flight, segments, audiences — has already been through this discipline. Briefs live in a brief template, not in the writer’s head. Audiences live in a segment definition, not in the email tool’s session memory. The prompt library is the last unsystematised artefact category most teams carry, and the fix is the same five layers everything else has already adopted.

What it earns you

A library that survives the hand-off, the new starter, the contractor on a one-week engagement. A team whose craft compounds against the file rather than evaporating with the chat session.

The clearest worked example is The Content Wrangler’s webinar-description workflow, in which a hundred webinar descriptions per year moved from manual writing to a modular template library composed in the Promptitude platform. Annual cost dropped from $25,000 to under $800 — about a hundred hours per year reclaimed (Promptitude case study, 2024 — vendor-published; the headline figure assumes a $250-per-hour fully-loaded rate, and the percentage shifts at a different rate). The case clears the case study standards and is logged in the subscriber-library register with the source URL and the rate caveat carried through.

The artefact stack scales the result. The full schema and a sixty-plus prompt-template library that follows it live in the Premium bundle; the prompt template schema page is public.


Liked this? Most of the work it draws on lives in the Playbook. See what's on offer or read two sample chapters.