claude

roles

Nine role agents for plan execution, from brief writing and candidate hunting to the final adversarial review and the phase orchestrator

0 on GitHubby Montinouv0.8.1MITupdated 4 days ago
/

Plugin roles from marketplace triqual in Montinou/triqual. Commands read from the repo's manifest — 4 days ago.

What it installs

Subagents (9)

  • brief-writer

    Extracts a task brief from an implementation plan plus the current code (interfaces, file paths, constraints, exact values). Reads and summarizes; makes no design decisions. Use to prepare the single source of requirements an implementer will work from.

  • complex-implementer

    Implements one plan task that the plan marks as complex (`**Role:** roles:complex-implementer`): multi-system changes, migrations of live state, or work where a mistake is costly. Works from a brief with strict TDD (write the failing test, run it, implement, run it green), verifies with focused test

  • final-adversarial-reviewer

    Single high-value adversarial review of a whole branch before merge: tries to break the implementation against the spec, hunts correctness and security defects across task boundaries, triages deferred and parked findings, and states what must be fixed before merge. Use once per branch, after all tas

  • hunter

    Hunts for defects in one diff package for one category (correctness, security, spec-drift or state) and writes its candidates to a single file. Finds and reports; never fixes, never grades severity. Use for the cheap, wide first pass of a review that a stronger reviewer then confirms or refutes.

  • phase-orchestrator

    Runs one phase of an implementation plan by following methodology:executing-plans for that phase only: dispatches the role agents task by task, runs the task scripts and reviews, and returns a short summary to the controller. Never edits product code itself. Use once per phase, in place of the contr

  • review-fixer

    Resolves review findings on a task branch: fixes what is real with the smallest correct change, and refutes with evidence what is wrong instead of implementing it blindly. Re-runs covering tests, commits, appends a fix report. Use for fix rounds after a task review.

  • safety-reviewer

    Mechanical checklist review of one task diff: commit message and trailers, secrets or credentials in the diff, forbidden paths or directories touched, file placement versus the plan's file structure, lint/typecheck evidence present, canary rules from the project's global constraints. Use for the che

  • spec-reviewer

    Judgment review of one task diff against its brief and spec: missing, extra or misunderstood requirements, and test honesty (tests assert real behavior, RED/GREEN evidence is genuine, edge cases covered). Traces code paths and branches. Use for the spec-compliance and quality gate of each task.

  • tdd-implementer

    Implements one plan task from a brief with strict TDD (write the failing test, run it, implement, run it green), verifies with focused tests and the typecheck/lint scoped to the touched files (never the full suite), commits, self-reviews and writes a report. Use for every implementation task in plan

View source on GitHub

More from Montinou/triqual

All 7 plugins →

Methodology core: triqual's own orchestrator, workflow skills that dispatch the roles agents and drive progress tracking (task ledger), global rules, triqual sync and doctor, drift notices; installs the other plugins

11 skills · 1 command · hooks

0

headjack

claude

Optional. A personal vault for agent skills: import the skills you already have, then search and install them only when a task needs them. Local mode runs on SQLite with no server (Node 22.16 or newer).

2 skills · 1 MCP server

0

Strict progress tracking: one entry per task, the overall status and per-plan task ledgers driven by triqual tasks, kept up to date by gates that enforce them and a hook that injects the state at every session start. Committed mode (progress.md and status.md in the repo), local mode (under .git) or store mode (a local database, nothing in the repo or its history).

1 skill · 1 command · hooks

0

Optional weekly local report on the health of the claude-mem memory (hygiene, storage and growth, Chroma health) plus LLM insights on proposed rules, skill changes and stale notes from allowlisted personal projects. Runs lazily from a SessionStart hook and shows one line until the report is read.

1 skill · hooks

0

Runs the triqual hooks in Codex through ~/.local/bin/triqual-hook.

hooks

0

Caps the agents a Workflow script launches per model: fable 4, opus 10, sonnet 15, haiku unlimited. Since 0.3.0 it also caps Agent and Task starts per model in a 120 s window

hooks

0

Maintain roles?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex plugins7 plugins