| name | forge-gap |
| description | Inventory the operator's workforce (agents, skills, plugins, installed MCP servers, recent research team SYNTHESIS files) and produce a ranked capability gap list. Diffs against VoltAgent's 10-category specialist taxonomy and wshobson/agents to find missing meta-orchestration, developer-experience, language-specialist, and quality-gate capabilities. Use when forge-lead starts a session without a specific gap in mind, when a retrospective in research-lead's MEMORY.md mentions a recurring capability hole, or when the operator asks "what capability are we missing". |
| disable-model-invocation | true |
| allowed-tools | Read Glob Grep Bash(gh api *) |
Capability gap detector
Inventory sources
Read these in parallel:
~/.claude/agents/**/*.md — flat agents + research/ team specialists. Extract frontmatter names, descriptions, tools.
~/.claude/skills/**/SKILL.md — personal skills. Extract frontmatter.
~/.claude/plugins/installed_plugins.json — installed plugins.
~/.claude/plugins/marketplaces/*/ — available-but-not-installed marketplace content.
~/.claude/teams/*/INDEX.md + ~/.claude/teams/*/*/SYNTHESIS.md — evidence from other teams' sessions of capability blockers.
~/.claude/agent-memory/*/MEMORY.md — distilled lessons mentioning missing capabilities.
Gap detection algorithm
- Build a tool-need set: for each agent file, extract the
allowed-tools or inferred tools from body prose.
- Build a capability-coverage set: for each skill, extract the description's primary capability (first 250 chars).
- Diff against taxonomy: for each VoltAgent category (01-core, 02-language, 03-infra, 04-qa-sec, 05-data-ai, 06-dev-exp, 07-domains, 08-biz, 09-meta-orch, 10-research), count how many specialists the local install has vs how many VoltAgent has. Categories with <30% coverage are major gaps; categories with 30-70% are medium; >=70% are covered.
- Cross-reference against in-flight sessions: read
~/.claude/teams/research/INDEX.md. Any gap whose owning team is actively being designed is deferred, not assigned to the Forge.
- Rank remaining gaps by: (a) count of agents that would use the new capability, (b) existence of a reference implementation in anthropics/skills or MCP Registry, (c) explicit mentions in research-lead's MEMORY.md.
Output format
Write to ~/.claude/forge/gap-reports/<timestamp>.md:
# Gap report — <timestamp>
## Major gaps (>30% coverage deficit vs VoltAgent)
1. **<category>** — <N> specialists missing. Examples: <list>.
- Reference implementation: <URL or "none found">
- Recommended action: install / adopt / author / defer-to-research
## Medium gaps (30-70% coverage)
...
## Already covered
...
## Recommended first-batch for this session
1. <gap> — <install|adopt|author|defer>, estimated 1-3 sub-skills
2. ...
Hard rules
- Never propose a gap that's being actively designed by another team (check INDEX.md first).
- Never propose building a capability that's already installed (double-check
installed_plugins.json).
- Rank by utility to the workforce, not by how interesting the gap is.