AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Pastiche Setup

skill-retz8-pastiche-pastiche-setup · by retz8

Use when bootstrapping pastiche for first-time adoption (post-`/pastiche-init`) — fills `KNOWLEDGE.md` and seeds `[GENERAL]` `WISDOM.md` rules section by section. Resumable across sessions; re-invoke to continue.

No reviews yet
0 installs
30 views
0.0% view→install

Install

$ agentstack add skill-retz8-pastiche-pastiche-setup

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-retz8-pastiche-pastiche-setup)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
3mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Pastiche Setup? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Pastiche — setup

Resumes from setup_progress in pastiche/config.yaml. Unit of work is one section per turn; within a section, draft candidates as a numbered list, confirm in natural language, write once after agreement.

Preflight

  1. Read pastiche/config.yaml. On parse failure: print the parser error verbatim and stop — "Fix the YAML and re-invoke."
  2. Read pastiche/FACT.md. If empty or missing, stop: "pastiche/FACT.md is empty or missing. Run /pastiche-sync (or re-run /pastiche-init) and re-invoke."
  3. Re-derive setup_progress from files (the file is the source of truth; config is the cache). For each KNOWLEDGE section: marker still present → stub, otherwise done. For general-wisdom: any [GENERAL] bullet present in pastiche/WISDOM.mddone, otherwise stub. Update config silently.

Lint

Every lint call below: shell pastiche-lint (on PATH via plugin bin/ — do not search for it), cwd = repo root. Forward stdout; on non-zero exit, also forward stderr. Don't revert; don't abort.

First run vs. returning run

If every setup_progress entry is stub, this is a first run.

First run:

  1. Greet (two lines): "Let's set up pastiche — we'll fill KNOWLEDGE.md with how your DS gets used, and seed a few system-wide rules in WISDOM.md. You can pause any time; I'll resume where we left off."
  2. DESIGN.md handshake. If design_md_reference is set in config, skip. Otherwise scan repo root for DESIGN.md:
  • Found → ask: "I found ./DESIGN.md. Want me to use it as reference? I'll draft candidates from it so you mostly review."
  • Yes → validate the path (resolve absolute, in-repo, readable, ≤200KB). On any failure, warn and proceed as if declined. On pass, write design_md_reference: ./DESIGN.md to config. Scan the body for ^## Overview or ^## Brand & Style; if either is present, copy that section's body into KNOWLEDGE.md's ## Brand Identity (replacing the placeholder).
  • No → leave design_md_reference: null.
  • Not found → one line: "No DESIGN.md at repo root — we'll go through this from scratch."
  1. Start with Brand Identity (below).

Returning run: skip greet and handshake. Print: "Picking up where we left off — ` of 13 done. Next up: `. Continue?" Then resume.

Section loop

Pick the next section: lowest stub in canonical order, or --section if provided.

After each section, prompt: "Next section: ``. Continue / switch / pause?"

  • Continue → next stub in canonical order.
  • Switch → ask which section; jump.
  • Pause → stop; adopter resumes by re-invoking.

If --all was passed, skip this prompt. Stop only when every stub is done (then run Wrap).

Brand Identity

Auto-ported case (the placeholder was replaced from DESIGN.md): show the ported prose verbatim. Offer keep / edit / rewrite.

  • keep → write + flip + lint.
  • edit → take natural-language patches, re-show, loop until keep.
  • rewrite → fall through to the freeform case.

Freeform case (no auto-port, or adopter chose rewrite): prompt:

> "Write or paste your Brand Identity prose — voice, posture, density, restraint, anything that matters for downstream design decisions. Length is your call; round-1 reads this in full on every task. > > Not sure where to start? Say "guide me" and I'll walk you through 4–5 prompts to draft it together."

  • If the adopter pastes prose, take it as-is, read back for confirmation, loop on natural-language edits until yes.
  • If they say "guide me" (or equivalent), ask 4–5 prompts in sequence: brand voice in 1–2 adjectives; density posture (dense vs. spacious, why); restraint signal (when the DS pulls back); the one rule that makes a screen feel off-brand; optional anti-example. Draft a prose paragraph from the answers, read back, loop.

Write the final prose into ## Brand Identity (replacing whatever was there). Flip setup_progress.brand-identity: done. Run lint.

Scenario sections (11)

Canonical order: Action buttons, Forms & input collection, Feedback & status, Overlays, Navigation & wayfinding, Content display, Layout & page structure, Date & time selection, Iconography, Visual hierarchy, Domain-specific patterns.

Per section:

  1. Draft candidates as a numbered list, blending two sources:
  • DESIGN.md (if design_md_reference is set): mine narrative for scenarios in this section's territory.
  • Canonical seeds (always; see Seed scenarios below).

Adapt to the adopter's FACT: drop seeds whose atoms aren't in FACT; add candidates suggested by FACT atoms that no canonical seed covered. Atom names backticked (FACT entries verbatim); prop expressions bare. Example: `Button variant="primary" `. Soft target 5–8; no hard cap, no padding.

  1. Confirm. Show the list. "Accept all, or tell me what to change (accept some / reject some / edit some / add new)?" Loop on natural-language responses until the adopter accepts.
  1. Write. Replace the section's _(empty — …)_ marker with the confirmed scenarios. Each scenario: one or more prose framing lines, then one or more lines.
  1. Flip + lint. Set setup_progress.: done in config. Run lint.
  1. Cadence prompt (unless --all).

[GENERAL] WISDOM phase (last)

Runs only when every KNOWLEDGE stub is done. Same shape as a scenario section; writes to pastiche/WISDOM.md; the unit is a tagged rule.

  1. Draft candidates, blending:
  • DESIGN.md (if set): mine posture / invariants, especially any "Do's and Don'ts"-shaped content.
  • Canonical seeds (4):
  • Tokens-only — never raw hex / px; always reference DS tokens.
  • Spacing scale discipline — no arbitrary gap / padding values.
  • Breakpoint discipline — no arbitrary media queries.
  • Accessibility floors — contrast, focus-visible, semantic HTML by default.

Adapt seed wording to the adopter's FACT token naming.

  1. Gate DESIGN.md-mined candidates. For each DESIGN.md-mined candidate, apply: (a) can it be tagged to specific FACT atom(s) instead? (b) does it hold for all UI in the project regardless of atoms used? On either failure, drop and note: "`` sounds scenario-conditional — re-route to KNOWLEDGE during the relevant section." Canonical seeds skip this gate.
  1. Show, confirm, write. Numbered list, natural-language response, loop until accepted. Write each accepted rule as - [GENERAL] . Append to WISDOM.md (after any existing content, before EOF). Flip setup_progress.general-wisdom: done. Run lint.

Wrap

After every section write, check setup_progress. If every entry is done, run the wrap on this turn (includes --section and --all invocations).

Print:

  1. Completion line, warm and brief: "Setup complete — pastiche is wired up for your DS."
  2. Light stats: sections done, total scenarios authored, [GENERAL] rules added.
  3. One concrete next step: "Invoke /pastiche on a frontend task — your loop will gate against KNOWLEDGE + WISDOM. As your DS grows, use /pastiche-write-knowledge and /pastiche-write-wisdom for incremental additions."

No emoji, no theatrics.

Mid-session degradation

If design_md_reference is set but DESIGN.md becomes unreadable during the session (deleted, renamed, exceeds 200KB cap), warn once: *"DESIGN.md at ` is no longer readable. Continuing without DESIGN.md context — candidates fall back to canonical seeds."* Do not clear designmdreference` in config. Proceed with canonical seeds for the rest of the session.

Flags

  • --section — jump to a specific section (revise a done section, or skip ahead). Wrap fires if this invocation lands the final flip.
  • --all — same flow, no continue / switch / pause prompt between sections.

Seed scenarios

  • Action buttons — primary CTA; destructive action; secondary / alternative path; loading state; icon-only; link-style.
  • Forms & input collection — text input; selection (radio / select); multi-select; date / time pick; file upload; inline validation; field grouping; submit / cancel.
  • Feedback & status — success confirmation (toast / inline); error (inline / banner / modal); warning; loading indicator; empty state; progress.
  • Overlays — modal (informational); modal (confirmation); modal (destructive confirmation); slide-over / drawer; popover; tooltip.
  • Navigation & wayfinding — primary nav; secondary nav; breadcrumb; tabs; pagination; back / cancel.
  • Content display — list (compact); list (rich); card grid; table; detail view; media (image / video).
  • Layout & page structure — page header; section header; sidebar; main / detail split; footer; container widths.
  • Date & time selection — single date pick; date range pick; time pick; relative date display.
  • Iconography — decorative icon; action icon (with affordance); status icon; brand mark.
  • Visual hierarchy — primary heading; section heading; emphasized text; muted / secondary text; numeric / metric display.
  • Domain-specific patterns — patterns unique to the adopter's product (e.g., refund flow, KYC step, dashboard widget). DESIGN.md drives this section heavily when present; canonical seeds are minimal.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.