Search skills/

Pro Workflow Claude skills

A Claude Code plugin with 41 skills and many hooks that stores your corrections in SQLite and adds searchable research wikis, quality gates and cost tracking.

rohitg00/pro-workflow 41 skills Planning, TDD & debuggingMemory & context
Our verdict Recommended with caveats

Interesting self-correcting memory idea, but the license is unclear and the hook code could not be reviewed.

2/5 Depth expertise shipped Note Security Review before installing 30d Maintenance Updated 30 days ago 2.9K Adoption stars ? License no clear license

What Pro Workflow does

Pro Workflow puts one SQLite store under your Claude Code sessions. When you correct the agent, it proposes a rule, you approve it, and the rule is saved, indexed for full-text search and loaded at the next session start. A second layer holds persistent research wikis on disk with a search index, which hooks can inject into prompts, and an auto-research loop can grow a wiki between sessions. The README also describes quality gates, git and secret guards, compaction-aware state and cost tracking with budget alerts.

The repo lists 41 skills averaging about 4 KB, 8 agents, 23 commands and 33 hook scripts across 22 events. Skills include agent-teams, batch-orchestration, context-optimizer, mcp-audit, token-efficiency, insights, and a few with scripts such as llm-council and survey-generator that call model APIs using your own keys. The hook scripts and store code were not in the snapshot we reviewed, and GitHub detects no license even though the README badge says MIT. Optional keys for embeddings and runners are set through the plugin configuration dialog. Installation is a plugin marketplace add, or the skills CLI for other agents.

Good for

  • Long-running Claude Code projects where you repeat the same corrections
  • Building a searchable local knowledge base on a topic
  • Users who want session cost and context analytics

Not for

  • Anyone who needs a clearly stated license
  • Minimal setups where 33 hooks are too much surface

What's inside depth 2/5

41 skills averaging 3.9 KB, a handful with scripts, no evals; much of the claimed hook and store code was not in our snapshot.

Skills
41
Reference docs
1
Scripts
10
Tests / evals
0
Templates
10
Avg SKILL.md
3.9 KB

Counted from the repository file tree. How we score depth.

Dependencies and costs

  • SQLite store managed by the plugin
  • Optional OpenAI, Anthropic, Voyage or other API keys for embeddings, council and survey scripts
  • Claude Code experimental agent teams flag for the agent-teams skill
  • Telemetry: none found; cost tracking is described as local.

Security review

Review before installing

A large Claude Code workflow plugin: 33 hooks (quality gates, secret scan on writes, git/commit guards, learning capture, wiki context injection on each prompt), a SQLite learnings store under ~/.pro-workflow, and optional LLM council, survey, embedding and skill-optimizer tools that use your own provider keys. Those keys go only to the matching provider, and the prompt-classifier option is off by default but would post prompt text to a local or Typesafe endpoint if enabled. Hooks run automatically on install, and the full script set was not read line by line.

Worth knowing

  • installs_hooksThe plugin documents (README: 33 hook scripts across 22 events) and registers about 33 command hooks in hooks/hooks.json that run automatically once installed: quality gates, secret scan, git guards, learning capture; some can block actions (exit 2 from secret-scan.js on Edit/Write, git guards on Bash).hooks/hooks.json
  • Needs an API key for its own serviceOptional plugin settings take Anthropic, OpenAI, OpenRouter, Fireworks, Voyage, GitHub, Typesafe and Laya keys; they are passed to a bundled MCP server and the council/survey/embedding/optimizer scripts, which send each key to its own provider's endpoint (a custom LLM_COUNCIL_BASE_URL gets only LLM_COUNCIL_API_KEY)..claude-plugin/plugin.json:42
  • Makes network requestsOptional calls: LLM providers and embeddings (api.anthropic.com, api.openai.com, api.voyageai.com, openrouter, fireworks), the 'system one' prompt classifier (off by default; local Laya URL or api.typesafe.ai) which posts the prompt text, and the research loop fetching DuckDuckGo, arXiv and GitHub.scripts/lib/system-one.js:11
  • Installs packages or toolsProvider-backed features need npm install and npm run build in the plugin folder (better-sqlite3 native module and @modelcontextprotocol/sdk, caret-ranged).package.json:34
  • Changes files outside your projectKeeps its SQLite learnings database and state under ~/.pro-workflow, appends to CLAUDE_ENV_FILE on directory change, and the skill optimizer can read and rewrite SKILL.md files in ~/.claude/skills when you run it.scripts/lib/system-one.js:29

What we checked

The repository (231 files) on Oct 29, 2026. Not a full review: we read 10 of 87 code files; it relies on code outside the repository: Most of the 87 code files, including the other hook scripts under scripts/, src/db, src/search, the optimizer trainer/llm modules, wiki skill scripts, council.js and build-survey.js, were checked with pattern searches and excerpts rather than read in full; npm packages better-sqlite3 and @modelcontextprotocol/sdk, and the dist/ build output (not in the repo).

NVIDIA SkillSpector static scan: 159 pattern hits (Agent Snooping 37, Tool Misuse 25, Data Exfiltration 23, Supply Chain 18, Excessive Agency 17, MCP Rug Pull 16). Most hits are documentation or test strings: safe-mode and deny-rule examples (git push --force, curl|sh, chmod 777) that the guard blocks, reference docs about ~/.claude/settings.json, and CLI cheat-sheet lines. The dependency advisories are for caret-ranged SDK/zod/body-parser versions in package-lock.json.

A review, not a certification. Skills change after we check them — read the files before installing.

License

GitHub detects no license; the README badge says MIT but no LICENSE file was in our snapshot, so confirm before commercial use.

Can it run unattended?

Yes. The README describes an auto-research loop that grows a wiki overnight, though we could not verify its stop conditions.

Alternatives to Pro Workflow

Superpowers

obra/superpowers

A 15-skill development methodology that makes coding agents clarify the goal, write a plan, work through it with subagents and verify before calling it done.

Recommended Check first Depth 3/5

Context Mode

mksglu/context-mode

MCP server plus hooks that keep bulky tool output out of the context window, index it in local SQLite, and restore session state after compaction.

With caveats Broad access Depth 3/5

Planning with Files

OthmanAdi/planning-with-files

Keeps task_plan.md, findings.md and progress.md on disk and re-injects them via hooks, so a plan survives /clear, compaction and crashes.

Recommended Broad access Depth 4/5

memsearch

zilliztech/memsearch

Cross-agent persistent memory for Claude Code, Codex, OpenClaw, OpenCode and DeepSeek Harness: Markdown files as the source of truth, with a Milvus search index.

With caveats Broad access Depth 3/5

FAQ

What is Pro Workflow?

It is a Claude Code plugin that saves your corrections as searchable rules in SQLite, keeps research wikis, and adds hooks for quality gates and cost tracking, with 41 skills and 8 agents.

How do I install Pro Workflow?

In Claude Code run /plugin marketplace add rohitg00/pro-workflow then /plugin install pro-workflow@pro-workflow. Other agents can use the skills CLI.

Is Pro Workflow safe?

We saw no hidden calls in the scripts we could read, but the hooks were not in our copy and the license is not detected by GitHub. Review the hooks before enabling them.

Sources: GitHub repository (stars, license, file tree, last push Sep 29, 2026). Review based on the README and SKILL.md files as published.