marketplace-dev
Convert a Claude Code skills repository into an official plugin marketplace so users
can install skills via claude plugin marketplace add and get auto-updates.
Input: a repo with skills/ directories containing SKILL.md files.
Output: .claude-plugin/marketplace.json + validated + installation-tested + PR-ready.
Route the task before editing
| User intent | Route |
|---|
| Create marketplace support for a repo that does not have it | Follow Phases 0–4 below |
| Add or update an ordinary plugin entry | Follow “Maintaining an existing marketplace,” then Phase 3 |
| Consolidate standalone skills into a suite, create a suite from existing skills, move a skill between suites, or rename a skill while moving it | Read references/suite_consolidation_workflow.md completely and follow it instead of the generic Phase 4 constraints |
| Reconcile installed caches after a merged suite migration | Invoke daymade-skill:skill-governance and use its post-migration workflow; do not mutate cache state here |
| Audit the whole repository across code, docs, security, PRs, issues, and manifest integrity | Invoke marketplace-health-check:marketplace-health-check only when the user requested a broad health check; it is not a prerequisite for a targeted suite migration |
Phase 0: Evidence Intake
Before editing an existing marketplace, collect evidence instead of relying on the
default template:
- Read the current
.claude-plugin/marketplace.json.
- Read this repo's marketplace rules (
CLAUDE.md, README install section, changelog).
- Read official docs for marketplace/plugin path semantics.
- Use the current conversation as evidence. Mine earlier local Claude Code sessions
only when the user explicitly asks for or approves that private source.
When that history source is approved, each project's sessions live under
~/.claude/projects/<escaped-cwd>/:
- Top-level files:
<session-id>.jsonl
- Subagent transcripts:
<session-id>/subagents/agent-*.jsonl
Useful search patterns (adjust keywords to the failure you are debugging):
grep -lc "marketplace.json\|claude plugin validate\|claude plugin install" \
~/.claude/projects/<escaped-cwd>/*.jsonl
grep -lc "Unrecognized key\|Plugin not found\|No manifest found\|Duplicate plugin" \
~/.claude/projects/<escaped-cwd>/*.jsonl \
~/.claude/projects/<escaped-cwd>/*/subagents/*.jsonl
Extract lessons as evidence-backed rules: command attempted, observed output, root
cause, final working command/config. Do not encode guesses from memory.
Phase 1: Analyze the Target Repo
Step 1: Discover all skills
# Find every SKILL.md
find <repo-path>/skills -name "SKILL.md" -type f 2>/dev/null
For each skill, extract from SKILL.md frontmatter:
name — the skill identifier
description — the ORIGINAL text, do NOT rewrite or translate
Step 2: Read the repo metadata
VERSION file (if exists) — this becomes metadata.version
README.md — understand the project, author info, categories
LICENSE — note the license type
- Git remotes — identify upstream vs fork (
git remote -v)
Step 3: Determine categories
Group skills by function. Categories are freeform strings. Good patterns:
business-diagnostics, content-creation, thinking-tools, utilities
developer-tools, productivity, documentation, security
Ask the user to confirm categories if grouping is ambiguous.
Step 4: Choose plugin boundaries
Claude Code has three separate levels:
marketplace -> plugin -> skill
- Marketplace name is used for install identity:
plugin@marketplace.
- Plugin name is the slash namespace:
/plugin-name:skill-name.
- Skill name comes from
SKILL.md frontmatter when the skill path points to a
directory containing SKILL.md directly.
Choose each plugin boundary by installation/update/cache intent:
- Single-skill plugin: use when the skill should install, update, and roll back
independently with a narrow cache.
- Suite plugin: use when related skills should share one namespace and one
install command, for example
/daymade-docs:mermaid-tools.
For detailed source/cache patterns and pitfalls, read
references/cache_and_source_patterns.md before changing source or skills.
Phase 2: Create marketplace.json
The official schema (memorize this)
Read references/marketplace_schema.md for the complete field reference.
Key rules that are NOT obvious from the docs:
$schema field is REJECTED by claude plugin validate. Do not include it.
metadata only has 3 valid fields: description, version, pluginRoot. Nothing else.
metadata.homepage does NOT exist — the validator accepts it silently but it's not in the spec.
metadata.version is the marketplace catalog version, NOT individual plugin versions.
It should match the repo's VERSION file (e.g., "2.3.0").
- Plugin entry
version is independent. For first-time marketplace registration, use "1.0.0".
strict: false is required when there's no plugin.json in the repo.
With strict: false, the marketplace entry IS the entire plugin definition.
Having BOTH strict: false AND a plugin.json with components causes a load failure.
source defines the installed plugin root. For single-skill plugins, point
source directly at the skill directory (e.g., "./tunnel-doctor") and omit
skills entirely — this is the official pattern used by 167/168 plugins in
anthropics/claude-plugins-official. For suite plugins, use
source: "./<suite>" with explicit skills array listing subdirectories.
Avoid source: "./" (installs full repo as cache) and skills: ["./"]
(rejected by Claude Code 2.1.x path-escape validator).
- Reserved marketplace names that CANNOT be used:
claude-code-marketplace,
claude-code-plugins, claude-plugins-official, anthropic-marketplace,
anthropic-plugins, agent-skills, knowledge-work-plugins, life-sciences.
tags vs keywords: Both are optional. In the current Claude Code source,
keywords is defined but never consumed in search. tags only has a UI effect
for the value "community-managed" (shows a label). Neither affects discovery.
The Discover tab searches only name + description + marketplaceName.
Include keywords for future-proofing but don't over-invest.
Generate the marketplace.json
Use this template, filling in from the analysis:
{
"name": "<marketplace-name>",
"owner": {
"name": "<github-org-or-username>"
},
"metadata": {
"description": "<one-line description of the marketplace>",
"version": "<from-VERSION-file-or-1.0.0>"
},
"plugins": [
{
"name": "<skill-name>",
"description": "<EXACT text from SKILL.md frontmatter, do NOT rewrite>",
"source": "./<skill-name>",
"strict": false,
"version": "1.0.0",
"category": "<category>",
"keywords": ["<relevant>", "<keywords>"]
}
]
}
Naming the marketplace
The name field is what users type after @ in install commands:
claude plugin install dbs@<marketplace-name>
Choose a name that is:
- Short and memorable
- kebab-case (lowercase, hyphens only)
- Related to the project identity, not generic
Description rules
- Use the ORIGINAL description from each SKILL.md frontmatter
- Do NOT translate, embellish, or "improve" descriptions
- If the repo's audience is Chinese, keep descriptions in Chinese
- If bilingual, use the first language in the SKILL.md description field
- The
metadata.description at marketplace level can be a new summary
Maintaining an existing marketplace
Use this section for ordinary entry additions or metadata updates. For a suite
consolidation, source relocation, suite membership transfer, or standalone-to-suite
migration, use the dedicated workflow linked in the routing table; those operations
have a wider breaking-change and documentation surface than an ordinary entry addition.
When adding a new plugin to an existing marketplace.json:
- Bump
metadata.version — this is the marketplace catalog version.
Follow semver: new plugin = minor bump, breaking change = major bump.
- Update
metadata.description — append the new skill's summary.
- Set new plugin
version to "1.0.0" — it's new to the marketplace.
- Bump existing plugin
version when its SKILL.md content changes.
Claude Code uses version to detect updates — same version = skip update.
Pair the bump with a CHANGELOG entry in the same commit; nothing checks it
for you, and the Pre-flight Checklist is the only checklist item that asks for it.
- Bump existing plugin
version when its source or skills changes.
The installed cache path and component resolution changed even if SKILL.md did not.
- Audit
metadata for invalid fields — metadata.homepage is a common
mistake (not in spec, silently ignored). Remove if found.
Adding a member skill to an existing suite
The most common edit, and the one that fails across the most CI rounds. Adding a
member (not a new standalone plugin) moves four things together:
- The suite's
skills array in marketplace.json lists the new member
(./skill-name). check_marketplace.py fails when a member dir with a SKILL.md
exists on disk but is absent from the array.
metadata.version strictly increases. A new member changes the suite's
layout signature (source + skills); check_version_progression.py rejects any
layout change without a metadata bump above base — rule 5 above (bump the plugin
version) is necessary but not sufficient here.
README.md and README.zh-CN.md each gain a ### **skill-name** detail
section. check_doc_skill_lists.py is a required check that extracts every
### **name** heading from both READMEs and fails if a marketplace skill is
missing (and flags the reverse as GHOST).
- The suite's invocation list (
/suite:skill-name) in both READMEs.
Every new file must also avoid absolute user paths (/Users/<name>/…) — the PII
guard blocks the push. Write ~/-relative and verify by re-running the real
detector, not by eye.
One further item has neither a hook nor a checker: the CHANGELOG entry.
check_version_progression.py only checks arrows whose destination matches what
the candidate ships, and the assertion it then makes is on the from end (a rebase
past another release moves it) — a wrong destination is silently skipped.
check_changelog_structure.py asserts that ## [Unreleased] appears exactly once and
is the first ## heading. Neither requires an entry to exist. A PR that bumps the
version and skips the changelog is therefore green everywhere and merges. The rule is
stated in marketplace-health-check/SKILL.md:90 ("flag it, don't merge without it"),
but the doc-PR path never routes there (marketplace-dev/SKILL.md:30), and the
broad-health-check route is itself scoped to "only when the user requested a broad
health check … not a prerequisite for a targeted suite migration"
(marketplace-dev/SKILL.md:33) — so on this path nothing asks for it.
Confirm the entry yourself, in the same commit as the bump.
The three checkers also take their base differently, and a usage error is not a verdict:
| command | base argument | a usage error gives |
|---|
check_version_progression.py | --base <ref> --candidate HEAD (see below) | exit 2 — argparse, a refused --candidate-index with nothing staged, or an unreadable base/manifest |
validate_changed_skills.sh | positional <base-ref> (defaults to origin/main) | exit 1, printing base ref '<x>' is unknown here |
check_marketplace.py | none | n/a |
So the code that means "my invocation was wrong" is per-script: exit 2 for
check_version_progression.py, exit 1 for validate_changed_skills.sh. In both cases
the repo was never judged — the same state as not having run it at all. Do not read
either code as "the checker found a problem": across the pair, exit 1 is a usage error
in validate_changed_skills.sh but a real finding in check_version_progression.py
— which is exactly why the printed message, not the number, is what you read.
The post_edit_sync_check hook only catches "SKILL.md edited but plugin version not
bumped". Items 2–4 have no hook, and the CHANGELOG entry has neither a hook nor a
checker — run check_version_progression.py and
check_doc_skill_lists.py locally before pushing rather than discovering them across
CI rounds. Pass --base <base-ref> --candidate HEAD. The script also accepts
--candidate-index, which reads the manifest blob out of the git index and diffs
staged paths only; with nothing staged that blob is HEAD's copy, so it re-verifies the
commit you already have and reports "no regression" about work it never saw. The green
is real about the index and silent about your edit.
The version gate in CI re-runs only when the PR head moves, and it resolves its base to
origin/$BASE_REF at run time. A PR whose base branch has since advanced therefore
keeps the green it earned against the older base, while its version number can already
be a regression against the current one. Re-run the gate against the current
origin/main immediately before merging, not only before pushing, and rebase if it
fails.
Phase 3: Validate
Step 1: One-shot pre-flight check
Run the bundled validator. It runs four checks in sequence and exits non-zero on
any required failure:
bash scripts/check_marketplace.sh # validates current repo
bash scripts/check_marketplace.sh /path # validates a target repo
What it checks:
| # | Check | Failure means |
|---|
| 1 | JSON syntax of .claude-plugin/marketplace.json | file is not parseable JSON |
| 2 | claude plugin validate . (skipped if claude CLI missing) | schema-level rejection (e.g. Unrecognized key: "$schema", duplicate names) |
| 3 | source + skills resolution for every plugin entry | a plugin entry points to a SKILL.md that does not exist on disk |
| 4 | Reverse sync (disk → manifest) | WARN-only: a SKILL.md on disk is not registered in any plugin entry |
Common schema failures and fixes:
Unrecognized key: "$schema" → remove the $schema field
Duplicate plugin name → ensure all names are unique
Path contains ".." → use ./ relative paths only
No manifest found in directory when validating an installed cache path → validate
the marketplace manifest or plugin source, not a strict: false cache directory.
Step 2: Installation test
# Add as local marketplace
claude plugin marketplace add .
# Install a plugin
claude plugin install <plugin-name>@<marketplace-name>
# Verify it appears
claude plugin list | grep <plugin-name>
# Check for updates (should say "already at latest")
claude plugin update <plugin-name>@<marketplace-name>
# Clean up
claude plugin uninstall <plugin-name>@<marketplace-name>
claude plugin marketplace remove <marketplace-name>
Step 3: Cache footprint test
After installation or update, inspect the actual cache. This is the only way to
confirm source produced the intended snapshot:
PLUGIN=<plugin-name>
MARKET=<marketplace-name>
CACHE=$(jq -r --arg id "$PLUGIN@$MARKET" '.plugins[$id][0].installPath' ~/.claude/plugins/installed_plugins.json)
find "$CACHE" -maxdepth 1 -mindepth 1 -exec basename {} \; | sort
Expected results:
- Single-skill plugin cache:
SKILL.md plus its own scripts/, references/,
assets/ as applicable.
- Suite plugin cache: only the suite member skill directories and suite-scoped
resources.
- If unrelated skill directories appear,
source is too broad.
- If cache entries are symlinks, the plugin is not self-contained; use canonical
source directories instead of symlink farms.
Step 4: GitHub installation test (if pushed)
# Test from GitHub (requires the branch to be pushed)
claude plugin marketplace add <github-user>/<repo>
claude plugin install <plugin-name>@<marketplace-name>
# Verify
claude plugin list | grep <plugin-name>
# Clean up
claude plugin uninstall <plugin-name>@<marketplace-name>
claude plugin marketplace remove <marketplace-name>
Pre-flight Checklist (MUST pass before proceeding to PR)
Run this checklist after every marketplace.json change. Do not skip items.
Automated checks
bash scripts/check_marketplace.sh
All four checks must pass. Treat the reverse-sync WARN as a real signal: an
unregistered SKILL.md on disk is almost always either an accidentally-dropped
skill you forgot to register, or dead code that should be removed.
Metadata check
Verify these by reading marketplace.json:
Per-plugin check
For each plugin entry:
Final validation
bash scripts/check_marketplace.sh
Must print RESULT: PASSED before creating a PR. A WARN [4/4] is acceptable
only when you have consciously decided to leave a SKILL.md unregistered.
Phase 4: Create PR
Principles
- Apply these pure-incremental constraints only when adding marketplace support to an
upstream repository. A user-approved suite consolidation necessarily moves existing
skill directories and updates the repository's install documentation; follow the
dedicated consolidation workflow for that route.
- Pure incremental (new-marketplace route only): do NOT modify existing skill files; update README only when needed to expose the new install path
- Squash commits: avoid binary bloat in git history from iterative changes
- Only add:
.claude-plugin/marketplace.json, optionally scripts/, optionally update README
README update (if appropriate)
Add the marketplace install method above existing install instructions:
## Install
 <!-- only if demo exists -->
**Claude Code plugin marketplace (one-click install, auto-update):**
\`\`\`bash
claude plugin marketplace add <owner>/<repo>
claude plugin install <skill>@<marketplace-name>
\`\`\`
PR description template
Include:
- What was added (marketplace.json with N skills, M categories)
- Install commands users will use after merge
- Design decisions (pure incremental, original descriptions, etc.)
- Validation evidence (
claude plugin validate . passed)
- Test plan (install commands to verify)
Bundled hooks (optional, auto-activated)
This skill ships two PostToolUse hooks under hooks/:
hooks/post_edit_validate.sh — runs claude plugin validate whenever a
marketplace.json file is written or edited.
hooks/post_edit_sync_check.sh — warns when a SKILL.md is edited but the
matching plugin entry in marketplace.json does not bump its version.
Both hooks are declared in this plugin's own manifest entry (plugins[].hooks),
so they activate automatically when the plugin is enabled in a Claude Code
session. No manual settings.json edit is required. To disable them, remove
the hooks block from this plugin entry in the user's installed copy or use
/plugin disable marketplace-dev (they take effect only when the plugin is
enabled).
These hooks are editor-time guardrails. They do NOT replace
scripts/check_marketplace.sh — always run the pre-flight check before a PR.
Anti-Patterns (things that went wrong and how to fix them)
Read references/anti_patterns.md for the full list of pitfalls discovered during
real marketplace development. These are NOT theoretical — every one was encountered
and debugged in production.