subagent-forge

Find and synthesize academic papers, preprints, and evidence for a topic. Use when the task needs current papers, methods, evidence trends, or source-grounded summaries.

Map ecosystems, directories, adjacent options, forks, and category clusters. Use proactively when the user wants the landscape, not just a shortlist.

Search Reddit, Hacker News, X, GitHub discussions, issue threads, and niche communities to extract what real users actually like, hate, or warn about. Use when the task benefits from raw community signal rather than polished summaries.

Verify the most current version, release status, API behavior, SDK docs, pricing, or product state for fast-moving tools and technologies. Use when stale knowledge would be risky.

Audit GitHub repositories for maintenance health, implementation quality, issue patterns, setup friction, security risks, and real-world viability. Use when choosing between repos, checking if abandoned, or figuring out if a repo is worth adopting.

Find and rank tools, apps, APIs, SDKs, libraries, and products with high-signal research, maintenance checks, price/value tradeoffs, and real user feedback. Use when the task is discovery, comparison, or recommendation.

Critique plans, outputs, layouts, codebase changes, or analyses to find weak reasoning, missing edge cases, flawed assumptions, or presentation problems. Use when the main thread needs a sharp second opinion.

Verify important claims using primary or near-primary sources, resolve contradictions, and label what is confirmed versus inferred or uncertain. Use when the main thread already has candidate findings and needs a factual backstop.

Save compact project-local lessons, prompt improvements, source blacklists/whitelists, and recurring failure modes after a task. Use when wrapping up a research task, "save what we learned", "log this lesson", or when a useful pattern emerges.

Score candidate tools, plugins, agents, products, or repos, keep the high-signal subset, and cut overlap, abandonment, or low-value options.

Turn a vague topic into an execution-ready research brief. Use when the user says "research this", "look into this", "what should we find out about", or provides an ambiguous topic that needs sharpening before subagents can work on it.

Use the standard quick, normal, deep-dive, and adversarial report formats for worker briefs and final synthesis.

Rank sources by credibility, freshness, firsthand evidence, maintenance health, and shill risk. Use when the user says "are these sources good", "rank these", "which sources should I trust", or when separating signal from SEO sludge.

Convert messy comparisons or findings into a clean spreadsheet plan, scoring model, and review checklist. Use when the user says "make a comparison table", "turn this into a spreadsheet", "score these options", or has raw findings that need structure.

Build compact, high-quality task packets for subagents with objective, scope, ranking criteria, stop rules, red flags, and expected report style.

1 hook across 1 event