Skip to content

SESSION-SUMMARY-2026-07-03.md

Latest commit

 

History

History
82 lines (70 loc) · 3.7 KB

SESSION-SUMMARY-2026-07-03.md

File metadata and controls

82 lines (70 loc) · 3.7 KB
Raw

Working Session Summary — July 3, 2026

Shipped v2.0.0 and v2.1.0 in one session: the library's largest restructure. This note is the handoff — what changed, what's open, and how to pick the work up.

What shipped (see CHANGELOG.md for full detail)

v2.0.0 — Pattern foundation and restructure:

  • Generative Guidance pattern rewritten as v2 (generative-guidance-pattern.md): 3 recommendations + "Other", standing bypasses (best guess, bulk drop), loop control, question budgets, confirm-before-build. v1 prompts are grandfathered: migrate on touch, never mass-rewrite.
  • interaction-modes.md: facilitation / checkpointed co-construction / autonomous investigation.
  • jinja2-prompt-structures.md: loops, switches, guards for loop/goal/agent-safe prompts.
  • Directory taxonomy: prompts/ (one-pass from context), prompt-generators/ (emit reusable prompts), workshops/ (guided session produces the artifact — split from prompt-generators), market-intelligence/ (autonomous research — renamed from a short-lived investigations/).
  • ~30 new assets; tooling in scripts/; generated catalog/.

v2.1.0 — Loops and strategy frameworks:

  • loops/: seven seasoned /goal /loop /batch /routine recipes from Dean's article "Prompts Aren't Dead. They Just Got a Bigger Vocabulary." Each runs at three levels (plain, loop lingo, Just Enough Jinja2) with canonical output formats embedded.
  • market-intelligence/ gained the strategy suite: TAM/SAM/SOM (moved from prompts/), battle-card-builder, SWOT, Porter's Five Forces, Ansoff — all evidence-cited investigation mode.

How to work in this repo now

  1. Read AGENTS.md first (authoring contract), then the pattern doc matching your prompt's mode.
  2. After any prompt edit: python3 scripts/validate-prompts.py (errors block) and python3 scripts/generate-catalog.py (regenerates catalog/, which is derived — never hand-edit).
  3. New prompts declare an interaction mode and follow Generative Guidance v2 if they facilitate. Include "When NOT to Use."
  4. Companion links use repo-relative paths (the validator checks them).

Open items (in priority order)

  1. Article URL: when "Prompts Aren't Dead" publishes, link it in loops/README.md and the root README (both currently cite the title only).
  2. Agent Strategy Canvas "Prompt 1 of 2": prompts/agent- strategy-canvas-prompt-template.md is labeled Part 2; Dean has Prompt 1 — add it alongside when provided.
  3. 79 validator warnings = the migrate-on-touch worklist: grandfathered files missing metadata fields. Clear a file's warnings whenever you edit it for any reason. Do not mass-fix.
  4. Storytelling generators are still on the v1 pattern — migrate on touch, same rule.
  5. Candidate future work: a literal (rendered) Jinja2 pipeline for agent builders — pairing the loops/ recipes with JSON schemas and a real renderer; a "customer discovery" cross-directory view in the catalog.

Gotchas for the next person

  • CLAUDE.md is gitignored in this repo — it exists locally on Dean's machine only. AGENTS.md is the committed equivalent and is current; if you use Claude Code, copy AGENTS.md's guidance or ask Dean for the local CLAUDE.md.
  • The catalog and this repo's counts (96 assets) regenerate from metadata comment blocks; if a count looks stale, re-run the catalog script rather than editing numbers.
  • resumes-resignations-reactions/, skeletons/, and flows/ were intentionally untouched in v2.x.

Session participants

Dean Peters, with Claude Code (Fable 5). Companion repos studied for this redesign: Product-Manager-Skills and Mitre-Skills (both local under ~/Code/).