purchasing

Apply one or more **buyer profiles** from `buyer-profiles/` to `spec.md` so the `/research` and `/recommend` phases bias toward a named purchasing pattern (BIFL, budget, minimalist, etc.).

Generate a focused comparison of 2–4 specific products without going through the full research/recommend flow. Useful when the user has narrowed down candidates and wants to scan them side by side.

Quick one-off evaluation of a single product against the current spec

Extract product information from screenshots, PDFs, and catalog images in `for-ai/` into a structured CSV at `data/extracted/products.csv`.

You are running the intake for this purchase. The goal is to populate `spec.md` at the repo root with the per-purchase brief.

Load the user's standing shopping preferences (country, currency, trusted/avoided brands, markup threshold, risk tolerance, preferred vendors, etc.) into the current session so they can be applied during `/intake`, `/research`, and `/recommend`.

Compare local market prices against international RRP with import-cost realism

List recent recommendations from the recommendations/ folder with their feedback status (liked, disliked, no-response, queued).

Record like/dislike/nuanced feedback against a past recommendation and update the preference profile so future suggestions improve.

User provides something they enjoyed (a film, book, podcast, video, course, etc.); the recommender suggests similar items they might also enjoy, grounded in the preference profile.

First-run setup for the recommendation workspace — picks format, captures initial preferences, builds the folder structure, and suggests MCPs.

Parse an OPML file (typically a podcast subscription export) from ingest/opml/ and use it to seed/update the preference profile.

Parse a reading list export (Goodreads, StoryGraph, Pocket, Instapaper, Readwise, etc.) from ingest/reading-list/ and use it to seed/update the preference profile.

Parse a watch history export (Netflix, Trakt, Letterboxd, JustWatch, etc.) from ingest/watch-history/ and use it to seed/update the preference profile.

Produce the final recommendation as a Typst-compiled PDF at `from-ai/report.pdf`.

Re-run the shortlist against the current spec and write a new dated recommendations file

Evaluate all candidate products against the spec and produce `from-ai/research.md`.

You are analyzing the user's hardware configuration to identify bottlenecks, compatibility issues, and upgrade opportunities.

You are helping the user compare specific components to make an informed purchasing decision.

You are creating a formal cost estimate document that can be used for internal approval, accountant review, or VAT deduction purposes.

You are documenting the hardware specifications for a machine. This profile will be used for compatibility analysis and upgrade recommendations.

You are generating specific component recommendations based on the user's hardware analysis, preferences, and budget.

You are conducting the initial user interview to set up this hardware planning workspace. This is a structured interview process to gather all context needed for effective hardware upgrade planning.

Run a one-time setup (or update) for the user's standing shopping preferences, and save them to memory (Mem0 MCP) so every future purchase workspace can `/load-preferences` them.

Update spec.md with an addition, edit, or removal — diff-only response

Researches manufacturer reputation, history, support quality, and regional presence. Use before recommending any product from an unfamiliar brand.

Interviews the user about their tastes, parses ingested history (OPML, watch history, reading lists), and maintains memory/preferences.md as the canonical preference profile.

Compares prices across vendors and markets, calculates markups, and factors in import costs (shipping, VAT, customs). Use when evaluating value across multiple sources or when deciding buy-local vs. import.

Extracts structured product data (name, brand, price, SKU, vendor) from screenshots, PDFs, and catalog images. Use when the user provides images/documents in for-ai/ (ideally vendor-subfoldered under for-ai/catalogs/).

Generates targeted content recommendations grounded in the user's preference profile and feedback history. Writes one markdown file per recommendation under recommendations/ with reasoning and metadata.

Gathers and synthesizes reviews from multiple platforms into a weighted score and a short honest summary. Use during research for each serious candidate.

Verifies claimed product specs against authoritative manufacturer sources. Use when a claimed spec is load-bearing for the recommendation (e.g., "has USB-C charging", "supports Wi-Fi 6E").

Provision a new purchasing workspace on disk. Use when the user wants to start a new purchase decision, tech-rig planning exercise, or vendor-recommendation workspace. Accepts a workspace name and optional variant (general-purchasing | tech-procurement | vendor-recommendations). Scaffolds the workspace, personalises CLAUDE.md from the user's global memory, and (by default) creates a GitHub repo.