using-bearpaws git:20261103.59550d9A

Immutable. This exact content never changes and is served at /api/v1/blob/380e63a099f5e21f.

---
name: using-bearpaws
description: Use when starting any conversation, before responding or exploring the project
---

<skill>

  <subagent-stop>If you were dispatched as a subagent to execute a specific task, skip this skill.</subagent-stop>

  <warning level="hard">
    If a skill might apply, invoke it before responding, asking, or exploring. Follow applicable skills; user instructions take precedence.
  </warning>

  <process>
    <step>**Claude Code:** Invoke `Skill` and follow it; do not `Read` skill files. Use the Claude Code capability mapping when a skill action overlaps a native capability.</step>
    <step>**In Antigravity:** Use native Agent Skills and the capability mapping.</step>
    <step>**Other Agent Skills agents:** Load `.agents/skills/` or `~/.agents/skills/` via the native tool (`skill` in OpenCode) or read `SKILL.md`. Users can type `$skill-name` (Codex) or `/skill-name` (Devin CLI).</step>
  </process>

  <see file="references/claude-code.md"/>

  ## Red Flags

  These thoughts mean STOP — you're rationalizing:

  | Thought | Reality |
  |---------|---------|
  | "This is just a simple question" | Questions are tasks. Check for skills. |
  | "I need more context first" | Skill check comes BEFORE clarifying questions. |
  | "Let me explore the codebase first" | Skills tell you HOW to explore. Check first. |
  | "Let me gather information first" | Skills tell you HOW to gather information. |
  | "I can check git/files quickly" | Files lack conversation context. Check for skills. |
  | "I'll just do this one thing first" | Check BEFORE doing anything. |
  | "This doesn't count as a task" | Action = task. Check for skills. |
  | "The skill is overkill" | Simple things become complex. Use it. |
  | "I remember this skill" | Skills evolve. Read current version. |
  | "This feels productive" | Undisciplined action wastes time. Skills prevent this. |
  | "I know what that means" | Knowing the concept ≠ using the skill. Invoke it. |
  | "I'll note that I skipped the review" | Disclosure is not review. Run it, or report INCOMPLETE. |
  | "I know where this is going" | Early confidence is not permission to skip inspection, planning, or verification. |

  ## Pace Control

  Check knowns, assumptions, and existing work before acting. Re-evaluate when scope or risk changes. Verify each required gate.

  **Momentum does not waive gates.**

  ## Skill priority

  1. **`bp:onboarding-to-a-project`** first for projects; skip only for abstract design with no project involved.
  2. **`bp:brainstorming`** next for design, using project conventions.
  3. **Other process skills** for planning, TDD, debugging, review, and verification.

  ## Elevated risk

  Size never lowers risk. Work is elevated-risk if it touches authentication or authorization, secrets or cryptography, money, data deletion or migration, untrusted input reaching paths, SQL, shell, or deserialization, concurrency, or a public API or schema. However small the change, elevated-risk work needs tests pinning the behavior that must not change, run and seen passing in a step before your edit; a failing test for the risk first (`bp:test-driven-development`); a diff limited to the fix, with no unrelated cleanup; an independent review (`bp:requesting-code-review`) before any completion claim; and a risk receipt with verification evidence (`bp:verification-before-completion`). Saying you skipped a step does not satisfy it. A defect the review finds in the same risk class as the task, in the code you changed, is part of the fix, not unrelated cleanup: fix it with its own failing test, or end `Status: INCOMPLETE` naming it. A defect is behavior a caller in this codebase or a user's input can reach: a negative or oversized amount is one. A guard for a type no caller passes (a NaN amount, a bool flag, a string timestamp) is not, and the review is not asked for one. Dispatch the review before your final report; if it has not returned, the report's status is INCOMPLETE, never done.

  **Risk receipt.** End every reply that changed files with `Risk: routine` or `Risk: elevated ([surface])`, `Verified:` (command and result), `Not checked:` (what you did not run, or none), and `Status: COMPLETE` or `Status: INCOMPLETE`. Elevated work adds `Pinned:` (tests passing before the edit), `Failing first:`, `Review:` (reviewer and verdict), `Open:` (each unfixed Critical or Important review finding, or `none`; any open finding means `Status: INCOMPLETE`), and `Assumptions:` (what you took as true without checking).

  ## Lazy-load contract

  `<see file="...">` and `<include ref="...">` are advisory. Read targets only when needed; do not load them automatically.

  ## Skill types

  Follow rigid process skills exactly; adapt flexible patterns to context.

  ## Brevity Policy

  - Answer directly and keep routine updates brief when no process skill sets the output format.
  - Follow a process skill's detail requirements when it applies.
  - Never compress safety warnings, migration risks, acceptance criteria, defect-preventing review findings, test failures and error output, or verification results.

</skill>