parcadei/continuous-claude-v3
6 skills6k installs22.8k starsGitHub
Install
npx skills add https://github.com/parcadei/continuous-claude-v3Skills in this repo
1Discovery InterviewDiscovery Interview is an agent skill that acts as a product discovery partner for solo and indie builders who have enthusiasm but not yet a buildable specification. Instead of jumping to implementation, the skill runs a disciplined interview: a short orientation on problem, audience, and whether the work is greenfield or incremental, then deeper passes that vary by project type so API, UI, CLI, mobile, and full-stack efforts each get relevant questions about data, UX, scaling, and integrations. The documented philosophy is explicit: do not ask obvious questions, do not accept shallow replies, and do not assume the user already knows the right technical choices. When gaps appear, the skill is instructed to educate and to research rather than guess. A written spec is only produced once understanding is complete, which makes the artifact suitable as input to planning skills or direct agent execution. Confidence is high for idea-stage and early validation workflows where scope churn is expensive.3.6kinstalls2Mathmath is a journey-wide agent skill that gives solo builders one conversational front door for symbolic computation, constraints, units, and lightweight math explanations inside continuous-claude-style workflows. It is for indie developers who want Claude Code or similar agents to stop guessing whether to use SymPy, Z3, or Pint when they say integrate, eigenvalues, or convert miles to km. The skill documents a routing table from user intent to the right script under .claude/scripts/cc_math, run via uv and Python, and it defers formal proof work to /prove. You can invoke it during Validate when sanity-checking pricing or models, during Build when implementing algorithms or tests, or during Grow when interpreting analytics formulas—any time a numeric or symbolic step blocks progress. It is procedural agent tooling, not a replacement for a computer algebra GUI or a dedicated data-science notebook pipeline.661installs3Github SearchGitHub Search is a thin agent skill that routes GitHub discovery through an MCP github server and a Python harness script. Solo and indie builders install it when they want the agent to look up implementations, repository metadata, issue threads, or PR history using native GitHub query syntax instead of manual browser tabs. Invocation is explicit: pass --type and --query, optionally scope with --owner and --repo, and let Bash execute the packaged script while Read handles any follow-up file inspection. The skill fits early idea research when you are surveying how others solved authentication or API patterns, during build when you are aligning with ecosystem conventions, and during ship or operate when you triage critical labels or regressions on upstream dependencies. It is not a full code-review or PR automation workflow—only search. Success depends on configuring the MCP server and a personal access token before the first query; without that, every run fails fast at configuration rather than returning partial results.631installs4PremortemPre-Mortem is a journey-wide agent skill for structured failure imagination: assume the project failed in three months and explain why, before you commit further effort. Solo builders invoke it via /premortem with auto depth, forced quick for plans and PRs, or deep before implementation, optionally scoped to a specific file. Risks are classified as Tigers (real threats), Paper Tigers (overblown), and Elephants (unspoken issues), following Shreyas Doshi's framing of Gary Klein's technique. A hard editorial rule requires verification—read surrounding context, confirm scope, and avoid flagging hardcoded paths without fallback checks—so output stays actionable rather than noisy. It pairs naturally with planning and review workflows in continuous-claude-v3 when you want agent-assisted risk passes without skipping human judgment on what is in scope.621installs5Dead CodeDead-code is an agent skill that tells your coding agent how to find unreachable and unused functions with TLDR static analysis instead of guessing from grep. Solo builders accumulating helpers and deprecated paths in SaaS, CLI, or API repos can invoke it before a refactor, during Ship review, or when Operate iteration feels bloated. The workflow documents `tldr dead .`, scoped directories, custom entry points like `main`, `cli`, `test_*`, and framework handlers, plus Python and TypeScript language flags. Output summarizes total versus dead function counts with file paths and line numbers so you can delete or quarantine safely. It intentionally trades automatic hook startup cost for manual, fast runs when you need them. Compatible with Claude Code and other agents that allow Bash per the skill metadata.552installs6Idempotent RedundancyIdempotent-redundancy is a compact methodology skill from the continuous-claude-v3 line that tells coding agents how to implement belt-and-suspenders reliability without creating write loops, conflicting updates, or repair storms. Solo builders shipping agents, integrations, or ops automation invoke it whenever they add fallback branches, reconciliation jobs, or secondary writers to the same state—Braintrust merges, filesystem checkpoints, and multi-step repair hooks are called out explicitly. The skill is user-invocable false, meaning it is meant to shape implementation discipline during agent work rather than as a standalone chat command. Apply it across Build when wiring integrations, Ship when reviewing defensive paths, and Operate when adding self-healing hooks. The outcome is redundant safety nets that behave the same on the tenth run as the first.1installs