harness-everything

Use at session start (Discover phase) to detect and align with the operating system, shell (Git Bash vs PowerShell vs CMD), package managers, and dev tools; prevents blind command execution and repetitive tool errors.

Evaluate AI agent performance across correctness, token efficiency, anti-loop focus, and environment awareness; save a 4-dimension scorecard. Use when asked to run a benchmark or score an execution log or conversation history.

Enforce fable-mode's context discipline as a shadow guard. Use when running staged multi-agent loops to prevent context bloat, cap per-stage output size, keep each stage within its physical token boundaries, and stop agents from reading files outside the current stage's scope.

Stage large, multi-source or multi-session tasks through a written plan, named fable agents, failable per-stage checks, and skeptical delivery review; use fable-opus, fable-sonnet, or fable-haiku for explicit model selection.

Search the agent-skills ecosystem (skills.sh / npx skills) when no installed skill fits; verify quality, apply ephemerally via temp cache by default, and install permanently via npx skills add only with explicit user approval

Generate Angular-style commit messages after verifying the environment, submodules, and staged files; use for explicit commit requests or concluded task phases.

Acts as a relentless challenger to stress-test plans and architectures by interrogating one question at a time, finding loopholes, combating AI sycophancy, updating the CONTEXT.md glossary in real time, and handing off resolved decisions to to-spec for formal specs/ADRs.

Challenge the plan against the existing domain model, sharpen terminology, and update CONTEXT.md and ADRs inline as decisions crystallise; use when stress-testing a plan against documented language and decisions.

Route software/project work through the Harness kernel: classify Tier 1/2/3, establish mandatory invariants, then let the agent choose tactics and domain skills. Use for software triage, routing, and re-routing; not general Q&A or non-software writing.

Discovers and transforms shallow or bloated modules into deep, SRP-compliant ones by splitting oversized objects/functions, extracting seams, and decoupling interfaces.

Explain or explicitly apply the Harness cognitive policy: Discover, Think, Try, Summarize, Record. It provides cross-cutting behavior without requiring this skill to be selected before domain skills.

Scaffold a verified multi-agent workspace and select bounded specialists from an external agency-agents catalog without vendoring the full external roster.

Generate or update precise, reader-oriented README.md and AGENTS.md based on actual project scans and smart merging of existing files. Use when a project lacks documentation or needs refreshed agent onboarding docs.

Rewrite past Git history to comply with Angular Style conventions. Use when cleaning unpushed commits before a PR, squashing fixups, rewording messages, or reordering history safely with abort on conflict.

Conducts STRIDE threat modeling, scans secrets via audit scripts, and hardens code against OWASP Top 10 risks with three-tier boundary controls. Use when implementing auth, handling input/uploads, managing secrets, or auditing code before deployment.

Extract root causes from resolved problems and persist them as defensive memory rules or dynamic skills; use after struggles, zoom-outs, or explicit request.

Author, audit, and refactor SKILL.md files against one quality bar — the Skill Contract format. Use when creating a new skill, refactoring a SKILL.md, checking overlap, or packaging an insight as a dynamic skill.

Apply Harness style guidelines when writing or refactoring SKILL.md files. Use when creating a new skill, reviewing skill structure, enforcing naming and frontmatter conventions, or standardizing tone, formatting, and progressive disclosure across the skill catalog.

Drive Standard Tasks (Tier 2) with Test-Driven Development (RED/GREEN/REFACTOR) backed by terminal test evidence. USE FOR: "implement a feature test-first", "fix a bug with a regression test", "refactor safely under tests". DO NOT USE FOR: "macro planning or scaffolding", "docs-only work with no testable logic".

Turn the conversation into a written spec — feature PRD, CLI/API reference, schema doc, or dev doc, whichever fits — and publish it per this repos projectDocs framework; use when alignment is complete.

Break a to-spec feature-spec, a plan, or a conversation into tracer-bullet tickets with declared blocking edges — one file per ticket locally or issues on a real tracker via projectDocs. Use when an approved spec needs dependency-aware ordering.

Enforce a deliberate step-by-step execution loop for complex multi-step or multi-file tasks (Tier 2/3), one sub-task in-progress at a time. USE FOR: "track a complex multi-step task as todos", "work through a multi-file refactor step by step". DO NOT USE FOR: "trivial single-file edits", "quick questions without implementation".

Use when starting feature work needing isolation or before implementation plans - ensures an isolated workspace via native tools or git worktree fallback

Run objective verification gates (build, types, lint, tests, security scan) before claiming work done. USE FOR: "run the full verification loop", "verify this is done before delivery". DO NOT USE FOR: "design discussions with no completed work", "planning without code changes".

Verify external framework/API claims and unmeasured perf/cost estimates against an authoritative source before asserting them. Use when stating how an external framework, SDK, CLI, or API behaves, or quoting numbers not produced by a real run; output is a cited claim, a real measurement, or a labeled estimate.

Reflect-first circuit breaker - after 3 failures, stop edits, rebuild the full picture with read-only tools, write a fact-check report, then resume or escalate. USE FOR: "you are stuck in a loop", "the same error keeps failing", "rethink your assumptions". DO NOT USE FOR: "routine single-failure debugging", "greenfield planning without failures".