agentleFS
hansohnhansohn/marketplace-skills/CLAUDE.md

marketplace-skills

CLAUDE.md1 starsChanged 4 months agoMITView on GitHub

A Claude Code plugin for preparing secondhand items for resale. It generates a complete listing file — pricing analysis, listing copy per enabled platform, and a recommended photo order — from an item name or a CSV row, and can optionally create an eBay draft. Scope: this file configures agents working on the plugin itself. The reusable assets are the skills in skills/. The plugin is named marketplace (in .claude-plugin/plugin.json), so its skills are invoked as /marketplace:<skill>. The on-disk folder…

The file

We split this CLAUDE.md into the rules it gives an agent

8 rules. About 0 are common, 0 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Skill anatomy: Every SKILL.md has YAML frontmatter (name matching the directory, description ≤1024 chars stating *what* + *when to use*), …

    line 43
  2. Very specific:

    skills/ holds only skills: Every directory under skills/ must contain a SKILL.md.

    line 48
  3. Very specific:

    Only fb-marketplace and ebay ship: If you add another, also add it back to the skill descriptions (they were trimmed to claim only what …

    line 59
  4. Very specific:

    Descriptions drive routing, not names: Keep each skill's description trigger-rich and distinct from its siblings.

    line 61
  5. Very specific:

    Always run validate-skills.js (and ideally the trigger evals + tests) after editing any SKILL.md.

    line 95
  6. Very specific:

    Never add a directory under skills/ that isn't a skill.

    line 97
  7. Very specific:

    Never hardcode a personal path or location in tracked files — .claude/settings.json is personal and untracked.

    line 98
  8. Very specific:

    Never let a skill's copy claim a platform that has no platforms/<name>.md.

    line 103

≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

marketplace

A Claude Code plugin for preparing secondhand items for resale. It generates a complete listing file — pricing analysis, listing copy per enabled platform, and a recommended photo order — from an item name or a CSV row, and can optionally create an eBay draft.

Scope: this file configures agents working on the plugin itself. The reusable assets are the skills in skills/. The plugin is named marketplace (in .claude-plugin/plugin.json), so its skills are invoked as /marketplace:<skill>. The on-disk folder and git repo are marketplace-skills — folder/repo names are intentionally independent of the plugin name.

Structure

.claude-plugin/plugin.json  → plugin manifest (name, version, description)
skills/<name>/SKILL.md       → one skill per directory (nothing else lives here)
agents/<name>.md             → least-privilege subagent personas (registered in plugin.json)
platforms/<name>.md          → one file per marketplace format (drives generation)
templates/listing.md         → output file skeleton with {PLACEHOLDER} slots
references/placeholders.md    → shared placeholder-source table (progressive disclosure)
references/locked-fields.md   → listing Step 3a deterministic-field rules (read at 3a)
references/pricing-method.md  → comps Step 4 deterministic pricing method (backs scripts/price.js)
docs/adr/                     → architecture decision records (why non-obvious calls were made)
scripts/                      → validation + eval tooling (zero-dependency Node)
evals/cases/<skill>.json      → per-skill routing datasets (run-trigger-evals.js)
evals/workspace/              → skill-creator eval run outputs (gitignored)
.claude/settings.json         → personal dev config (untracked; see settings.example.json)

Skills

  • listing — orchestrator. Gathers item details, delegates to comps and photos, renders one copy block per enabled platform, writes {ItemPrefix}-Listing.md, and optionally chains into ebay-draft.
  • comps — pricing research only (live eBay sold data via Playwright, optional FB).
  • photos — reviews a photo folder, returns a ranked order + condition notes.
  • ebay-draft — pushes a prepared listing to eBay as a draft via the eBay MCP.

Conventions

  • Skill anatomy. Every SKILL.md has YAML frontmatter (name matching the directory, description ≤1024 chars stating what + when to use), then these sections in order: When to use → Required inputs → Process → [Output format] → Verification → Notes. The Verification section is a checklist of the skill's hard invariants (e.g. ebay-draft's "never publish").
  • skills/ holds only skills. Every directory under skills/ must contain a SKILL.md. Eval outputs, snapshots, and scratch do not belong here — that is what let the old listing-workspace/ pollute discovery. validate-skills.js fails on any non-skill directory here.
  • Plugin-root paths. platforms/, templates/, references/, and sibling skills live at the plugin root, not inside a skill folder. Resolve them as ${CLAUDE_PLUGIN_ROOT}/platforms/… etc. A bare platforms/foo.md read relative to skills/listing/ will 404.
  • Platforms are file-driven and extensible. The orchestrator iterates platforms/*.md; adding a marketplace is just a new file (sections: ## Aliases, ## Format rules, ## Block template, ## Generation rules) — no code change. Only fb-marketplace and ebay ship. If you add another, also add it back to the skill descriptions (they were trimmed to claim only what ships).
  • Descriptions drive routing, not names. Keep each skill's description trigger-rich and distinct from its siblings; validate-skills.js warns on overlap and run-trigger-evals.js measures routing.

Key invariants (enforced in each skill's Verification section)

  • Never auto-publish or auto-post anywhere. ebay-draft creates a draft and stops; the user publishes from Seller Hub. listing writes a file; the user posts.
  • No surviving placeholders in a generated listing ({...} / [YOUR_CITY]).
  • Live pricing only — comps pulls live data and never fabricates prices.
  • Location resolves to a real user-supplied value, never a placeholder city.
  • Surface ambiguity with options, never guess. On missing, conflicting, or ambiguous input (location, platform, condition, price, includes, which item), state the confusion, give concrete labeled options, and ask — don't invent a value or split the difference. See references/handling-ambiguity.md.

Tooling

node scripts/validate-skills.js         # structure/frontmatter check (CI gate)
node scripts/check-location-leaks.js    # pre-publish PII guard: fails on any real "City, ST ZIP"
node scripts/run-trigger-evals.js       # routing regression tripwire (--min-rank1 80)
node scripts/price.js <prices...>       # deterministic resale pricing calculator (used by the comps skill)
node --test scripts/validate-skills-test.js scripts/run-trigger-evals-test.js scripts/check-location-leaks-test.js scripts/price-test.js

run-trigger-evals.js scores each labeled query in evals/cases/<skill>.json against all skill descriptions with a TF-IDF lexical proxy — it catches a description edit that tanks a skill's discriminability; it is not a substitute for real-model routing. Behavioral (execution) evals for listing live under skills/listing/evals/ and run via the skill-creator plugin.

Boundaries

  • Always run validate-skills.js (and ideally the trigger evals + tests) after editing any SKILL.md.
  • Never add a directory under skills/ that isn't a skill.
  • Never hardcode a personal path or location in tracked files — .claude/settings.json is personal and untracked; ship changes via settings.example.json. The only location string allowed in committed files is the placeholder Phoenix, AZ 85016; check-location-leaks.js scans the whole publishable tree (tracked + untracked-not-ignored) and fails on any other City, ST ZIP. Run it before publishing.
  • Never let a skill's copy claim a platform that has no platforms/<name>.md.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.