Samuel

Plugin samuel from marketplace samuel-skills in samosunaz/agent-skills. Commands read from the repo's manifest — 2 days ago.

Skills (39)

• address-pr-comments

  Triage, verify, and resolve incoming PR comments — apply fixes, reply, resolve threads, close each pass with a Resolution marker comment, re-request review. Incremental: scopes to feedback since the last pass. Author side of the PR review gate. Trigger on 'address pr comments', 'address review feedb

• analyze

  Read-only cross-artifact consistency check across spec, plan, tasks, journal, and constitution. Severity-tagged findings, no auto-edits. Trigger on 'analyze', 'consistency check', before implementing a non-trivial feature.

• api-contract

  Document an endpoint the backend built, for the client(s) that consume it — web frontend and/or native mobile (iOS/Android). Emits TS + Swift/Kotlin types. Output is an inline copy-paste block by default. Trigger on 'api contract', 'document endpoint', 'contract for frontend/mobile'.

• api-request

  Spec an endpoint a client needs from the backend, from the web frontend OR native mobile (iOS/Android) perspective. Output is an inline copy-paste block by default. Trigger on 'api request', 'i need an endpoint', 'request endpoint', 'spec for backend', 'new endpoint from front/mobile'.

• biz-tldr

  Retell a technical text (a review, plan, PR, or set of findings) for the person who owns the product: a headline, the flow end to end as who-does-what steps, what can go wrong and what it costs someone, internal-only fixes in one line, and the owner's decisions with a recommendation. Trigger on 'biz

• cascade

  Run one item down the intelligence tiers and back up: this session (the owner's flagship model) frames the problem and answers the planner's checkpoints, a high-tier planner writes the Executor Plan, a medium-tier implementer executes it, a fresh high-tier reviewer audits the diff, and this session

• conductor

  Drive the pipeline unattended phase-by-phase. Can bootstrap from an item id, drive setup→…→validate, and (in --ship mode) open a draft PR. Can be fired automatically by GitHub (schedule / issue-labeled) — see reference/automated-trigger.md. Trigger on 'conductor', 'run pipeline', 'run the pipeline u

• coordinate

  Coordinator mode for one task: own direction, decomposition and sign-off; hand every routine implementation to named Orca workers (Claude or Codex) launched with an explicit model and effort, a six-slot brief and a compact report contract; supervise without silent waits; integrate and verify in one

• create-atomic-commit

  Create git commits with user approval and no AI attribution. Reviews changes, plans atomic commits following conventional commits, and executes upon confirmation.

• create-constitution

  Ratify the repo's CONSTITUTION.md (3-5 non-negotiable principles, semver, governance) at v1.0.0. Optional — the pipeline degrades gracefully without it. Trigger on 'create constitution', 'ratify principles', 'initial constitution'.

• create-review-md

  Generate a repo's schema-v1 REVIEW.md — deterministic evidence digest (repo type, gate signals, CI job names, validation/journal findings, self-review history) + semantic derivation of repo-specific rules, cited per bullet. Existing REVIEW.md → --check conformance report only. Trigger on 'create rev

• debrief

  Report the outcome of work that already landed — what it achieved in the owner's terms, whether any of it is visible and where to look, what was measured, and what comes next in the epic. Reads merged PRs and their issues; writes nothing. Trigger on 'debrief', 'what did we gain with this wave', 'tld

• done

  Close the loop — open a PR (Closes #N), mark the item done, and clean up. Supports --draft for autonomous runs. Trigger on 'done', 'finished', 'ship it', 'open the PR'.

• feature-dossier

  Create or update a living feature dossier — enriched markdown + Mermaid diagrams/flows — in a versioned product catalog. For big features that change product behavior; reference for humans and AI before future modifications. Output language configurable (--lang, default es). Trigger on 'feature doss

• find-unknowns

  Map-vs-territory audit — surface known unknowns, unknown knowns, and unknown unknowns of an idea, Issue, or diff before they get expensive. Modes: audit (default), preflight (Issue N), teach (--teach «domain»), quiz (--quiz [N]). Standalone — no task-context required. Trigger on 'find unknowns', 'un

• iaas

  Drive one item through Implement → [Audit → Address] × N → Simplify as chained headless phases, each with fresh context. The round ceiling comes from the plan's size chip or --rounds. Trigger on 'iaas', 'audit loop', 'run the audit rounds', 'implement audit address simplify'.

• implement

  Execute a self-contained Executor Plan step-by-step with human verification between phases, keeping a living Implementation Notes journal. Use when ready to implement a planned item.

• interrogate

  The agent's own first-principles pass before it reports work as done — self-invoked at every close-out (implement Step 3, validate Step 2, iaas Simplify, coordinate C6), rarely typed by a human: restate what the change is for, then challenge every piece against it — what is unnecessary, over-complic

• kickoff

  Initialize a new indie project: vision doc, MVP scope, tech decisions, initial task breakdown. Trigger on 'kickoff', 'new project', 'start project'.

• land

  Land a set of finished pull requests as one train: read each PR's real state from the host, order them by file overlap, merge the whole set behind ONE approval, re-verify what each merge invalidates, then sweep the branches, worktrees and panes they leave behind. Trigger on 'land', 'land these', 'me

• mermaid

  Mermaid diagram style standard: semantic shapes + emoji vocabulary + classDef palette. Invoke before authoring any Mermaid diagram (dossier, contract, RFC, Issue/PR). Trigger on 'mermaid', 'flow diagram', 'diagram this', 'mermaid standard'.

• motion-brief

  Refine a natural-language description of a desired animation, transition, or motion effect — optionally with screenshots, diagrams, GIFs, or reference videos — into a precise, implementation-ready Motion Brief using canonical animation nomenclature. Use when the user describes motion loosely ("the c

• next

  Pull the next prioritized work item (GitHub Issues). Trigger on 'next', 'next task', 'what is next', 'pull task'.

• plan

  Create a self-contained plan (human Brief + Executor Plan) through 5 interactive phases with forced human checkpoints. Writes to the GitHub Issue body. Use when planning a feature, refactor, or initiative before implementation.

• polish

  Run the whole simplification chain over a finished branch as ONE run — `/samuel:interrogate` (delete what should not exist) → the runtime's native `/simplify` (tidy what stays) → `/samuel:remove-slop` (strip the slop) — then re-run the project's real checks and report once. Takes several PRs as a bl

• pr-self-audit

  Review a PR for bugs, security, and logic errors. High-signal only. Trigger on 'review pr', 'code review', 'audit this pr'.

• premise

  Burn a product fact agents keep re-deriving wrongly into the file they actually read, then sweep the repository for everything that contradicts it. Fires when the human corrects the same assumption a second time. Trigger on 'premise', 'burn this in', 'write this down where agents read it', 'you keep

• remove-slop

  Remove AI-generated code slop from the current branch. Scans diff against base branch for unnecessary comments, defensive code, type hacks, and style inconsistencies introduced by AI tools.

• repo-audit

  Substrate drift detector for samuel consumer repos: deterministic checks (samuel.md, gh auth/repo, pipeline+roadmap+promo labels, .claude gitignored, CLAUDE.md; optional: constitution, REVIEW.md, conductor heartbeat, release-please, squash merge, TL;DR templates, cross-session messaging) + a semanti

• roadmap

  Product Ownership session — read the product's real state and propose prioritized bets (what to build next and why), persisted as roadmap:* issues ready to promote into the pipeline. Discovery, upstream of plan. Trigger on 'roadmap', 'product direction', 'what to build next', 'what is next for the p

• session-handoff

  Create or resume handoff documents for context compaction between sessions. Use when context is large, switching sessions, or resuming previous work.

• spec

  Generate a Spec doc (User Stories P1/P2/P3, FR-### MUST, SC-### measurable, Edge Cases) with up to 5 clarifications. Optional — only when spec_required. Trigger on 'create spec', 'write the spec', or after /samuel:start-task when spec_required.

• start-task

  Pick a work item (GitHub Issue), create a branch, and inject task context. Use when a dev is ready to start implementing an item.

• tdd

  Write tests at the seams the plan declared, one behaviour per red-green cycle. Refuses to invent a seam the human never confirmed. Trigger on 'tdd', 'testing seams', 'write the tests'.

• tldr

  Prose standard: rewrite a message to Simplified Technical English — one reading per sentence, no filler, no abstract jargon. Targets the last assistant message by default, or pasted text / a file path. Never touches code, commands, or quoted output. Trigger on 'tldr', 'bro', 'rewrite that', 'make it

• update-constitution

  Amend the repo's CONSTITUTION.md. Semver versioning (PATCH / MINOR / MAJOR), Sync Impact Report, updates Last Amended. Reads existing file at repo root, proposes surgical edits, applies after user approval.

• validate

  Validate the implementation against the plan, verify acceptance criteria, run the project gate, resolve & seal the journal, and analyze documentation impact. Mandatory final step before done.

• wave-prep

  Prepare the open backlog for a wave run: sweep pipeline issues, infer real inter-issue dependencies from their Executor Plans, declare the missing native blockedBy edges (human-approved), and hand the ready set to /samuel:waves. Trigger on 'wave prep', 'prepare waves', 'prepare the backlog for waves

• waves

  Attended multi-issue coordinator: executes a set of pipeline:ready issues in parallel waves computed from GitHub's native blockedBy graph — one Orca worktree + worker per issue (Codex default, claude-conductor variant), draft PRs, human merge releases the next wave. Trigger on 'waves', 'launch waves

Subagents (7)

• cascade-implementer

  Claude implementer of /samuel:cascade, selected with `--medium sonnet:high` (size S) — executes a ratified Executor Plan in its own worktree with the least code that meets it, runs the real checks of every project it touched, and reports in six lines. Stops and asks on a plan–reality mismatch.

• cascade-planner

  Planner tier of /samuel:cascade — turns the coordinator's framing into a Brief + Executor Plan grounded in the code, with measured claims and prior-art checks. Changes no code. Returns open questions instead of guessing.

• cascade-reviewer

  Review tier of /samuel:cascade — a fresh-context adversarial audit of one frozen commit against the issue's acceptance criteria and the ratified plan. Validates every finding by executing it, leaves the tree as it found it.

• component-locator

  Locates files, directories, and components relevant to a feature or task. Call `component-locator` with human language prompt describing what you're looking for. Basically a \"Super Grep/Glob/LS tool\" — Use it if you find yourself desiring to use one of these tools more than once.

• implementation-analyzer

  Analyzes codebase implementation details. Call the implementation-analyzer agent when you need to find detailed information about specific components. As always, the more detailed your request prompt, the better!

• implementation-reviewer

  Adversarially reviews a code change against its spec/acceptance criteria before it ships. Call to independently verify an implementation — finds bugs, security issues, unmet criteria, and vacuous tests — and return a structured verdict (APPROVE | REQUEST CHANGES). Unlike the locator/analyzer agents,

• pattern-scanner

  Finds similar implementations, usage examples, or existing patterns that can be modeled after. Gives you concrete code examples based on what you're looking for. Like component-locator but also returns code details.

Hooks (1)

• PreCompact

  sh "${CLAUDE_PLUGIN_ROOT}/scripts/precompact-handoff.sh"

Similar plugins