SPECTRE

The second major feature group is the session memory and handoff system, which maintains and accumulates context across multiple coding sessions. By turning off auto-compact in Claude Code settings and running /spectre:handoff when the context window is getting full, users can save a status report that is automatically loaded into the next session's context window. If previous sessions exist, a subagent reviews the last three status updates and merges them into a single continuous session memory, providing trailing three-session memory snapshots. This system allows for seamless continuation of work, prevents loss of progress, and enables users to multi-task effectively, as they can start fresh with /spectre:forget when switching gears, ensuring that the agent remains aware of the ongoing thread of work without manual context management.

A third critical capability is the evaluate phase, which combines architecture review with knowledge capture to build institutional memory. Running /spectre:evaluate dispatches a subagent to produce a principal-level architecture review of completed work in parallel with capturing durable project knowledge—such as patterns, gotchas, decisions, features, and procedures—into reusable skills. These skills auto-load in future sessions via a SessionStart hook that reads the project's knowledge registry and injects relevant information into context, and skill auto-loading triggers when task keywords match registered skills. This creates a compounding knowledge loop where each session builds upon previous learnings, making future sessions faster and more accurate as the agent proactively knows what it has already documented and learned.

The product works overall through a series of specialized slash commands that correspond to each phase of the SPECTRE acronym: Scope, Plan, Execute, Clean, Test, Rebase, and Evaluate. Users start with a kickoff prompt like /spectre:scope, and the agent guides them through subsequent steps automatically, suggesting next steps at the end of each phase. The workflow leverages subagents for specialized tasks such as implementation, analysis, finding code, pattern recognition, web research, testing, and independent code review, though users do not need to know these subagents exist as the prompts instruct Claude Code to call them automatically. Technical implementation involves saving canonical docs to a structured directory, maintaining session logs, and using hooks for memory and skill loading, all designed to integrate seamlessly with Claude Code and Codex environments.

Key benefits and measurable outcomes for users include achieving higher quality and more consistent results from coding agents, enabling agents to work autonomously for much longer durations, and significantly increasing typical output—potentially by 10 to 100 times—in a repeatable manner. Users can ship product features faster across various project types, including websites, React Native apps, native desktop apps, and personal software, regardless of codebase size. The workflow reduces the cognitive load on developers by providing clear next steps and structured processes, minimizes time spent on manual specification writing, and enhances code quality through automated cleanup, testing, and knowledge retention, ultimately making complex product development feel easier and more manageable.

Concrete use cases involve building new features from scratch, such as scoping a feature with /spectre:scope, defining user flows with /spectre:ux, creating a technical design with /spectre:plan, and executing tasks with /spectre:execute, which includes parallel subagent work and validation. For low-complexity tasks, users can employ /spectre:ship to autonomously handle everything from brain dump to pull request. Debugging sessions utilize /spectre:fix for structured investigation and fixes, while ongoing maintenance involves /spectre:sweep for light cleanup and /spectre:clean for deeper code quality checks. The workflow also supports rebasing with conflict handling via /spectre:rebase and final evaluation with /spectre:evaluate to capture learnings, demonstrating adaptability across the entire development lifecycle.

The target users are product builders, developers, and technical product managers who use AI coding agents like Claude Code, seeking to enhance their productivity and code quality. Integrations are primarily with Claude Code via a plugin installation and with Codex using an npm command, supporting both user and project scope installations. The tech stack includes the plugin architecture for Claude Code, subagents for specialized tasks, hooks for session memory, and skills for knowledge capture, all managed within a GitHub repository structure. Pricing plans are not explicitly mentioned, but the product is open-source under an MIT license, suggesting free usage with community support through issues, email, and social media channels like Threads, X, and LinkedIn.

In summary, SPECTRE provides a comprehensive, structured workflow that transforms how developers interact with AI coding agents by systematically eliminating ambiguity through canonical documents, maintaining session continuity, and compounding knowledge across projects. Its primary value lies in making high-quality, agent-assisted product development repeatable and scalable, enabling users to tackle complex features with confidence and significantly amplify their output without sacrificing code quality or alignment with their vision.