viber

Plugin viber from marketplace p2p2 in p2p2sp/p2p2.claude. Commands read from the repo's manifest — 30 days ago.

Skills (18)

• code-auditor

  Multi-agent code audit through one lens (bugs, security, web-performance, runtime-performance, tests or design) over one scope (the current diff, one directory or the whole repository), every finding verified by an independent agent. Use when the user asks to audit, review or hunt defects in code fo

• commit

  Use whenever the user wants to commit, save, or record changes to git - including "commit", "commit changes", "commit all", "commit these files". This is the ONLY path to a commit: never run git add/commit yourself, never branch, never inspect git status/diff first - a forked agent stages, writes th

• create-issue

  Creates a GitHub issue from the project's issue form templates. Use when the user asks to open an issue, not to fix a bug.

• create-pr

  Opens a pull request for the current branch. Only on the user's request.

• e2e

  Generates and locally verifies Playwright tests for the QA scenarios of one viber run, then commits them. Use it after a build closed with a qa.e2e.md handoff file in its run directory.

• extension

  Creates or adapts an agent in the project's .claude/agents/ that a viber build runs at its close, just before the run is archived, with one forked skill in .claude/skills/ per phase when its work runs in phases one after another, and registers it as an entry of build.extensions in .claude/viber.yml.

• fixer

  Traces a reported bug to its root cause and proves it with a failing test, then hands the diagnosis to the planner - it applies no fix itself. Use whenever user reports a bug and wants to fix it. Not for a fix the user asked to apply directly, without a plan.

• handoff

  Writes a handoff file that lets a fresh session continue the current conversation - where to look, what is done, the decisions with the alternatives they rejected, what comes next and the open problems. Saved under .temp/viber/handoff/ unless the first argument is a directory or .md path (it holds a

• help

  Opens the viber usage guide in the default browser.

• implementor

  Builds an approved plan task by task. Requires an existing plan; without one, suggest the viber:intent interview.

• intent

  Planning interview - asks what the conversation and the code leave open, one question at a time, sizes the scope, proposes the spec shape and hands a confirmed summary to viber:planner. Never started unprompted. When a change looks like it needs a plan and no confirmed interview or viber:fixer diagn

• memory

  Reviews and repairs the host project's CLAUDE.md cascade - maps every node with its own size and the size of the chain a reader loads with it, creates the nodes a project with none needs, verifies each existing node against the code of the area it describes, brings every node below the root within b

• planner

  Only for a confirmed viber:intent interview or a viber:fixer diagnosis already in context - never the entry point. Without one, suggest the viber:intent interview and let the user decide. Turns that input into a reviewed implementation plan - acceptance criteria, file map, then tasks carrying depend

• prototype

  HTML mockup of a UI change in the project's own look, one proposal or three variants. Use when the user wants to see or compare a UI change before planning it.

• rules

  Reviews and repairs the host project's .claude/rules/ directory - maps every rule with its own size, the scope it declares and the number of tracked files that scope really matches, proposes the rules a scope with none needs, verifies each existing rule against the code it gates, and resets the laye

• setup

  Prepares a project for viber - seeds .claude/viber.yml, .gitignore and the recommended .claude/settings.json permissions (asking whether to merge into or reset an existing one that differs from them), checks the gh CLI is installed, checks whether CLAUDE.md names the build and test commands, and ope

• talk

  Research conversation on one topic - a free exchange of views in which Claude takes and defends positions, challenges weak ones and has every claim that carries weight checked against trusted sources on the web through viber:researcher, ending on conclusions with their sources, never on code, a plan

• triage

  Reviews one reported issue against the current code - whether it can be done, how, what it affects and how big the change is - classifies it and names the next viber step without taking it. Takes a GitHub issue number or URL, or pasted issue text, and can post the report as a comment on the issue.

Subagents (24)

• CLAUDE
• adr-screener

  Screens the decisions of one plan against the ADR admission test and returns the records worth writing, and the records to deprecate or extend. Invoked only by the planner skill, never directly.

• arbiter

  Rules on one stalled point of a build in the user's place - chooses one option from a closed list, given the case and its reports, and returns it with a one-line reason and a one-line cost if wrong. Invoked only by the implementor skill, never directly.

• closeout

  Records where a finished build delivered something the run's specification does not promise, accounts for every file the build left uncommitted, writes the build summary, then archives the run directory. Invoked only by the implementor skill, never directly.

• critic

  Independent verifier for a single hunter finding. Invoked only by the code-auditor skill, never directly.

• e2e-writer

  Turns one scenario of a run's QA handoff file into one Playwright spec and proves it green against the running application. Invoked only by the e2e skill, never directly.

• final-reviewer

  Reviews one slice of a finished build past the diff of its own tasks, or rechecks the fix of the build's final review, and writes a findings report on failure. Invoked only by the implementor skill, never directly.

• hunter

  Frontier depth-first investigator of one unit or one lens angle. Invoked only by the code-auditor skill, never directly.

• mapper

  Lens-aware repository mapper - reads the target's memory, history and the lens's map signals and writes the run's map of units worth a look. Invoked only by the code-auditor skill, never directly.

• memory-auditor

  Verifies one CLAUDE.md node against the code of the area it describes. Invoked only by the memory skill, never directly.

• memory-node-writer

  Writes one CLAUDE.md node per dispatch within the node budget - corrects an existing node from its audit findings, or authors a node from its own area's code. Invoked only by the memory skill, never directly.

• memory-writer

  Folds what a finished build taught into the project's CLAUDE.md nodes. Invoked only by the implementor skill, never directly.

• plain-plan-review

  Reviews one plan written in plain plan mode and returns PASS or FAIL with blocking findings. Invoked only on the plan gate's request, never directly.

• planner-review

  Reviews one implementation plan and returns PASS or FAIL with blocking findings. Invoked only by the planner skill, never directly.

• prototype-writer

  Writes, revises or narrows one self-contained HTML mockup of a UI change in the host project's own look - a single proposal or three variants behind an A/B/C switcher. Invoked only by the prototype skill, never directly.

• prover

  Verifies every claim of one drafted interview question - its recommendation and alternatives - against the code and, for facts outside the repository, the web, and returns CONFIRMED or REVISED. Invoked only by the intent skill, never directly.

• qa-writer

  Turns a finished build into the run's QA scenarios - the acceptance document a person performs by hand and the handoff file the e2e run automates. Invoked only by the implementor skill, never directly.

• researcher

  Checks a batch of claims from one research conversation against trusted sources on the web and, for claims about the repository, against the code, and returns a verdict per claim with its sources. Invoked only by the talk skill, never directly.

• rules-auditor

  Verifies one rule file under .claude/rules/ against the tracked files its paths glob matches, or proposes a candidate rule for a directory with none, every candidate passed through the admission gate at rule-admission.md. Invoked only by the rules skill, never directly.

• rules-writer

  Folds the conventions a finished build confirmed, or a rules review found, into .claude/rules/. Invoked only by the implementor skill and the rules skill, never directly.

• scout

  Cheap breadth-first triage scorer of mapped units. Invoked only by the code-auditor skill, never directly.

• task-coder

  Implements one plan task, or fixes one review report, and proves it green. Invoked only by the implementor skill, never directly.

• task-reviewer

  Gates one implemented plan task against its definition and writes a findings report on failure. Invoked only by the implementor skill, never directly.

• test-runner

  Runs the project's build, its fast tests and the integration tests a change reaches once and returns a verdict, keeping the log out of the caller's context. Invoked only by the implementor skill and the intent skill's fast path, never directly.

Hooks (6)

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/plan-gate.sh"

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/kill-guard.sh"

• PreToolUse

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/shell-guard.sh"

• UserPromptSubmit

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/plan-hints.sh"

• SessionStart

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/session-start.sh"

• SubagentStop

  bash "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/verdict-nudge.sh"