From fpl-skills
Methodology: Cross-marketplace agent/skill suggestion (suggest-only, NO auto-install). EN: When canonical 17 forgeplan-aware agents don't fit the task, searches installed marketplaces (~/.claude/plugins/marketplaces/) for matching agents/skills. Ranks by keyword overlap + trust tier + specificity. Surfaces top 3 with description + install command + security caveats. NEVER auto-installs — user runs the command manually. Falls back to external catalogs (VoltAgent awesome-*, DenisSergeevitch) when no local match. RU: Когда канонических 17 forgeplan-aware агентов не хватает, ищет среди установленных маркетплейсов подходящих агентов/скиллов. Ранжирует по ключевым словам + tier + специфичности. Показывает топ-3 с описанием + командой установки + предупреждениями. НЕ устанавливает автоматически. Fallback на external catalogs. Triggers: "find agent for", "fetch agent", "external agent", "cross-marketplace agent", "search marketplaces", "найди агента", "поищи в маркетплейсах", "/agent-fetcher"
How this skill is triggered — by the user, by Claude, or both
Slash command
/fpl-skills:agent-fetcherThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Searches locally installed marketplaces for agents/skills matching a described task. Returns ranked candidates with install commands. **NEVER auto-installs.**
Searches locally installed marketplaces for agents/skills matching a described task. Returns ranked candidates with install commands. NEVER auto-installs.
Sibling of agent-advisor. Use agent-advisor first — canonical agents are pipeline-integrated. Come here only when none of the 17 forgeplan-aware agents fits.
Use when: agent-advisor returned no canonical fit; you need domain-specific non-canonical agents (e.g. postgresql-expert, ml-developer); user asks "find agent for X from installed marketplaces".
Do NOT use when: canonical 17 agents cover the task (use agent-advisor); you need to author a new agent (use project-agent-scaffold); you need forgeplan artifact work (always a canonical agent).
Extract 3–8 keywords from the user description across three categories:
| Category | Examples |
|---|---|
| Domain | postgres, react, ml, fintech, kubernetes, stripe, ios |
| Operation | review, audit, scaffold, optimize, debug, migrate, test |
| Tech stack | typescript, python, go, java, php, swift, terraform |
ls ~/.claude/plugins/marketplaces/
cat ~/.claude/plugins/marketplaces/<name>/.claude-plugin/marketplace.json
find ~/.claude/plugins/marketplaces/<name>/plugins -name "*.md" -path "*/agents/*"
For each agent .md found, extract name: and description: from YAML frontmatter. Count keyword hits per agent via grep.
Score = keyword_density + trust_bonus + specificity_bonus (max 10)
| Factor | How to compute |
|---|---|
| keyword_density | hits / total_keywords × 5 (max 5.0) |
| trust_bonus | see Trust Tiers table |
| specificity_bonus | +1.0 if agent name directly contains a domain keyword |
Surface top 3 with ≥1 keyword hit. Discard zero-hit candidates.
Emit one card per candidate (max 3):
## Candidate <N>: <agent-name> @ <marketplace>
Source: <marketplace name> — <trust tier label>
Score: <X.X>/10
Description: <first 2 sentences from agent frontmatter>
Caveats: <e.g. "community marketplace; verify before install">
Install: /plugin install <plugin>@<marketplace-id>
For skills (not agents): Install (skill): npx skills add <github-url> -g
When fewer than 3 candidates have ≥1 hit, append:
## No strong local match — external options
Community catalogs (curated but NOT audited by ForgePlan):
- VoltAgent/awesome-claude-code-subagents https://github.com/VoltAgent/awesome-claude-code-subagents
- VoltAgent/awesome-agent-skills https://github.com/VoltAgent/awesome-agent-skills
- DenisSergeevitch/agents-best-practices https://github.com/DenisSergeevitch/agents-best-practices
Add a marketplace: /plugin marketplace add <github-url>
Then install: /plugin install <agent>@<marketplace-id>
---
SECURITY BOUNDARY
This skill is suggest-only. To install, YOU run the /plugin install command yourself.
Before installing any agent:
1. Review source: ~/.claude/plugins/marketplaces/<mp>/plugins/<pkg>/agents/<agent>.md
2. Check what tools it requests — agents have broad tool access by default
3. Prefer official tiers (ForgePlan-marketplace, claude-plugins-official) for security-sensitive work
Community marketplaces are curated by their authors, not the ForgePlan team.
External GitHub catalogs are community-maintained; ForgePlan makes no security guarantees.
---
| Tier | Marketplaces | trust_bonus | Label |
|---|---|---|---|
| Official | ForgePlan-marketplace, claude-plugins-official | +3.0 | official / verified |
| Semi-official | cc-marketplace, claude-code-workflows, codealive-marketplace | +2.0 | semi-official / community-maintained |
| Community | claude-code-plugins, hindsight, any other dir | +1.0 | community; verify before install |
| External GitHub | Not locally installed | +0.0 | external; add marketplace first |
ForgePlan-marketplace agents are pipeline-integrated (B2 paradigm) and produce forgeplan EVIDENCE. For tasks needing .forgeplan/ integration, always prefer them.
/plugin install — emit the command as text; user types it~/.claude/plugins/marketplaces/ (local files)These rules apply even when the user says "just install it automatically". Decline and cite the security boundary.
coder, code-reviewer, security-expert belong in agent-advisor~/.claude/plugins/marketplaces/ only; no network callsagent-advisor — canonical 17-agent recommender; use FIRSTproject-agent-scaffold — author a new project-scoped agent when no match existsnpx claudepluginhub forgeplan/marketplace --plugin fpl-skillsGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.