thedotmack/claude-mem
16 skills27.4k installs1M starsGitHub
Install
npx skills add https://github.com/thedotmack/claude-memSkills in this repo
1Mem Searchmem-search is an agent skill for querying claude-mem’s persistent memory database across sessions. Solo builders shipping with long-running Claude projects lose thread when they ask whether a bug was already fixed, how authentication was implemented last week, or what decision was recorded in an earlier sprint. The skill enforces a three-step discipline—search for a compact index with IDs, filter to the relevant observations or sessions, then fetch only those rows—so agents do not burn tokens loading full history blindly. Parameters cover project scoping, observation types such as bugfix and decision, and date ranges. It is most natural during Build and agent-tooling work but equally valuable in Ship review and Operate iterate when you need provenance for a regression. It does not replace reading the current conversation; it answers what happened in prior sessions. Requires claude-mem MCP tools configured for the agent environment.3.9kinstalls2Smart Exploresmart-explore teaches coding agents a token-efficient codebase navigation ritual backed by tree-sitter AST parsing and Claude-mem MCP tools. While active, the skill is explicit that smart_search, smart_outline, and smart_unfold should replace the usual Glob-then-Grep-then-Read discovery loop. Solo builders shipping with Claude Code or Cursor benefit when onboarding to a large repo, hunting shutdown hooks, or locating functions without dumping entire files into context. The workflow is deliberately three layers: discover files and symbols in a path, get a structural outline for one file, then unfold only the symbol you need to edit. It pairs naturally with implementation and review passes where structure matters more than line-by-line scrolling.3.2kinstalls3Make Planmake-plan is an orchestrator skill for creating detailed, phased implementation plans with embedded documentation discovery. Use it when tasked with planning a feature, API integration, or multi-step implementation — especially before executing with a "do" tool. It enforces a delegation model where subagents gather concrete facts (API signatures, file paths, examples) and the orchestrator synthesizes phase boundaries and task framing. This approach prevents implementing against outdated or assumed APIs and ensures each phase can be executed independently in new chat contexts.3.2kinstalls4DoDo Plan is a claude-mem workflow skill for solo builders who already have a phased implementation plan—especially one from make-plan—and want the main agent to act as foreman, not lone implementer. The protocol assigns one clear objective per subagent and demands evidence: commands run, outputs, files touched. Implementation subagents must copy patterns from documentation, cite unfamiliar APIs in comments, and halt when an API cannot be verified. After each phase, separate subagents run verification checklists, grep for documented anti-patterns, and perform code quality review; commits deploy only when verification passes. The orchestrator confirms completion against the plan before advancing. This reduces context rot on large features and mirrors how small teams parallelize review and QA. Tagged multi-phase because verification and commit gates span Build execution and Ship testing/review habits. Intermediate complexity: you need a trustworthy plan and tolerance for multi-subagent latency.3.1kinstalls5Timeline ReportTimeline Report is an agent skill for projects already instrumented with claude-mem. When you want a timeline report, a “journey into [project]” story, or a full development-history summary, the skill drives a structured pull from claude-mem’s persistent memory timeline and assembles it into a long-form narrative rather than scattered session notes. Solo and indie builders using Claude Code benefit when they need credible project archaeology—what was tried, when priorities shifted, and how features evolved—without manually replaying dozens of chats. Prerequisites are explicit: the claude-mem worker must be running and the project must have observations stored. The workflow begins by resolving the worker port from CLAUDE_MEM_WORKER_PORT, ~/.claude-mem/settings.json, or the per-UID default, then querying the memory API consistently. It is not a generic markdown exporter; it is tightly coupled to the claude-mem ecosystem and best used when memory coverage is already good.2.8kinstalls6Claude Code Plugin Releaseclaude-code-plugin-release is an agent skill that automates semantic versioning and the full release path for Claude Code plugins. It is aimed at solo and indie builders who maintain plugins in the mem/claude-mem ecosystem and need manifest files, npm packages, and marketplace listings to stay in lockstep. The workflow insists you draft detailed release notes before touching versions, then analyzes change impact, increments semver consistently across package.json, marketplace.json, and plugin.json, verifies builds, publishes to npm, tags git, opens GitHub releases, and emits changelog output from release JSON. Critical guardrails treat uncommitted build artifacts as release blockers and require a final git status check so nothing is left local-only. Use it when you are ready to ship a plugin version—not for exploratory refactors or partial manifest edits.2.2kinstalls7Knowledge AgentKnowledge Agent is a claude-mem skill for solo and indie builders who already accumulate observations and want reusable expertise instead of re-searching chat every time. You define a corpus with build_corpus—filters include project, observation types, concept tags, file prefixes, semantic query, ISO date bounds, and a limit that defaults to 500 observations—then prime_corpus to load that knowledge into an AI session before asking questions conversationally. The mental model is a focused brain: everything about hooks, all decisions from the last month, or bugfixes for one service. It fits when users ask to create knowledge bases from observation history, query past work patterns, or compile topic expertise. Pair it with normal agent workflows in Build and Operate so institutional memory stays queryable without a separate wiki.1.9kinstalls8PathfinderPathfinder is a journey-wide architecture orchestration skill for solo builders facing messy or grown codebases. When you are asked to find the ideal path, unify duplicated systems, or audit architecture before a refactor, the agent acts as orchestrator: subagents discover files, trace flows, and diagram features while the main agent synthesizes boundaries, unification strategy, and the final unified flowchart. Outputs land under PATHFINDER-<date>/ with feature inventory, per-feature flowcharts, duplication findings, and prompts for /make-plan per system. It fits Ship review before cutting scope for launch, Build backend when integrations proliferated, and Operate iterate when production cruft slows fixes. The skill enforces citations and redeploys weak subagent reports. Advanced complexity due to multi-agent coordination and Mermaid discipline.1.4kinstalls9Learn Codebaselearn-codebase is a claude-mem agent skill that forces a disciplined, exhaustive read of an entire codebase before meaningful implementation work. Solo builders invoke it when taking over a fork, joining a legacy service, or starting agent-assisted development on a repo they have not internalized—matching user phrases like learn the codebase, read the codebase, prime, or get up to speed. The procedure requires every source file to be read in full, using chunked reads with offset and limit when files are large so nothing is skimmed away at the end. It is intentionally token-heavy upfront to build a durable context cache that pays back across later planning, debugging, and review turns. Use it with Claude Code or similar agents when wrong assumptions about module boundaries would be more expensive than one thorough orientation pass; skip it when the change is trivial and localized and you already have an approved spec for that file.1.3kinstalls10How It WorksHow It Works is the claude-mem explainer skill for solo builders using Claude Code who want a precise mental model before enabling persistent memory. It describes the observation pipeline from routine tool use through session summaries and selective reinjection into later prompts, so you stop re-explaining architecture and rediscovering decisions every morning. The skill clarifies the second-session injection rule, the optional learn-codebase pass for dense repos, and the on-disk layout under the home directory. That matters for ship and operate concerns too: you need to know what leaves the machine only when compression providers are configured. Invoke it when someone asks what the extension is doing in the background or whether project memory is safe to turn on for a client repo.1.2kinstalls11Babysitbabysit is an agent skill for staying with a pull request through the full review and CI cycle until it is genuinely ready to merge. Solo builders and small teams invoke it when they want the agent to monitor, babysit, or keep checking PR comments, reviews, and checks rather than declaring victory after one gh pr view. The workflow identifies PR number and branches, rejects stopping on draft or dirty merge state, polls status checks, reads new human and bot feedback with skepticism toward summary bots, fixes verified issues in focused commits with local tests, pushes, and loops. GitHub CLI and GraphQL patterns are documented for coarse and fine-grained status. It pairs naturally with ship-phase testing and launch prep but belongs on review because the stopping condition is social and technical cleanliness on the PR itself.1.1kinstalls12WowerpointWowerpoint is an agent skill that pipes exactly one narrative source document through NotebookLM to produce a kawaii-styled slide-deck PDF. It is built for solo and indie builders who already have (or can draft) a comprehensive long-form doc and want a presentable deck without manually designing slides. Triggers include phrases like “wowerpoint this,” “make a deck about this file,” or “turn this report into slides.” The workflow covers one-time tooling setup with uv-installed notebooklm-py, Playwright Chromium, and jq, then an auth pre-flight before generation. The skill enforces quality by refusing to compensate for weak sources with extra inputs—you write or enrich the single doc first. Output stays slide-deck only; other NotebookLM media types are intentionally out of scope so expectations stay honest.835installs13Weekly DigestsWeekly Digests converts a full claude-mem project timeline into a serial, week-by-week narrative where each chapter represents one ISO week and reads the prior week's summary for continuity. Solo builders use it to understand how a project evolved through its complete history in digestible weekly stories rather than one long report. This matters for retrospectives, knowledge transfer, and maintaining a clear narrative arc of what was built, decided, and learned over time.415installs14Design IsDesign Is is an orchestrator skill for solo builders who have a concrete UI or design artifact and want a principled critique before planning changes. It walks through Dieter Rams’ ten principles in a fixed order, scores each from 0 to 3 with mandatory evidence, and synthesizes a clear verdict—whether you need a net-new design, targeted refinement, or a full redesign. The skill is built for agent workflows: you stay in evaluation mode until scoring is complete, then you receive a structured handoff prompt for `/make-plan` so planning stays separate from auditing. It complements but does not replace code review (`/review`), copy-only edits, or brainstorming when nothing exists yet. Best when you can point at real screens, components, or copy and want design quality framed as usefulness, aesthetics, and restraint—not subjective hot takes.414installs15Oh My Issuesoh-my-issues is an agent skill for solo and small-team maintainers drowning in GitHub noise. It treats symptoms as data and architectural defects as the real unit of work: you cluster related reports, assign each cluster a plan-master issue and matching design doc under plans/, redirect duplicates with a consistent comment template, and land one bundled PR that closes the cluster atomically. Invoke when you have roughly twenty or more open issues that feel like platform-specific duplicates, when someone asks you to triage, dedupe, consolidate, or derive a plan series from the backlog, or when a fresh bug clearly belongs on an existing master. The workflow compounds because fixes retire whole families of reports, prevention lands in CI via the plan test matrix, and standardized triage keeps new inflow cheap. It pairs naturally with agent-driven planning and PR workflows rather than ad-hoc one-off issue closes.414installs16StandupStandup is an agent skill that turns parallel coding branches into a single conversational room backed by one markdown file. Each branch-aware agent reads the file’s goal and prompt first, catches up on new messages, and contributes one substantive turn per invocation—introducing honest branch state early, then debating overlaps, risks, and merge strategy with @mentions. A human or lead agent facilitator drives round order and completion, which keeps individual agents from spinning on scheduling. The CLI surface (read/post/status/agree) is intentionally minimal so the behavior stays dialogue-shaped rather than form-filling. For solo builders running multiple Claude Code branches on the same feature, it offers a structured way to converge on what to keep, what to drop, and how to collapse worktrees before you merge or ship.1installs