All skills
simota avatar

/builder

@c805268
by shingo imotasimota/agent-skills85 stars
15

Implementing robust business logic, API integrations, data models, and reproducible AI image-generation code with type safety. Use for production implementation, Gemini image API pipelines, or interactive pair programming.

Use this Skill: https://skilld.dev/gh/simota/agent-skills/builder

This session only. Nothing lands on disk.

referencepair-programming.md

≈1.3k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Pair Programming Mode

Interactive, increment-by-increment co-implementation where Builder acts as the driver (writes production-grade code) and the user acts as the navigator (sets direction, reviews each increment, decides go/adjust/redirect). Unlike batch build (implement the whole feature, present once), pair programming is a conversation: propose one small increment, agree on it, implement it, verify it, confirm, then advance.

When to Use

Use pair programming Use batch build (default)
Live session, want to steer as it is built Spec is locked, want the feature delivered
Learning / mentoring / unfamiliar domain Well-understood change
High-uncertainty design — decisions emerge as you go Clear plan, low ambiguity
User explicitly asks to "pair", "build together", or "confirm as we go" Standard implementation request

Not a speed mode. For throwaway rapid prototyping (speed over quality), use Forge. Builder-pair keeps the full production quality bar — it changes cadence, not standards.

Roles: Driver / Navigator Separation

  • Builder = Driver. Proposes the next increment, writes the code, runs its verification, shows the diff. Owns how it is built to the quality bar.
  • User = Navigator. Owns what and whether — sets direction, approves each increment, redirects. Decides pace.
  • The navigator is not a passive reviewer. Builder proposes intent + verification before writing, so the user steers before code exists, not after.

Unlike Judge's pair-review (where Judge stays report-only and spawns a distinct driver to preserve generator ≠ evaluator), Builder is the generator — pair mode simply makes its build loop interactive and human-steered.

Protocol

SETUP:
  - Agree on the goal + acceptance criteria.
  - Draft an ordered increment plan (smallest shippable units). A Sherpa decomposition can seed this.
  - Agree on the verification path for increment #1 (test / type / contract / expected output) — verification-first.

LOOP until goal met or user ends session:
  1. Builder proposes the NEXT increment (one type module / one function / one vertical slice):
     - what it will implement, why, and the verification that will prove it
  2. User decides: go / adjust scope / redirect / skip
  3. Builder implements THAT increment only — to the full Core Contract quality bar
  4. Builder shows the diff + runs the increment's verification, reports the result
  5. User confirms (accept) or requests a change (bounded iterate — max 2 turns/increment)
  6. Checkpoint. Advance to the next increment.

CLOSE:
  - Run the 5-axis Impact Scope Check (callers / tests / types / configs / docs).
  - Present a session summary + handoff (Radar for tests, Guardian for PR).

Interaction Contract

  • One increment at a time. Never implement the whole feature then ask for a single approval — the point is steer-as-you-go.
  • Intent before code. State what + why + verification, get the go-ahead, then write. The navigator steers before code exists.
  • Increment size bounded. Each increment is reviewable in one sitting (≈ one slice / a small handful of functions). If an increment balloons, split it.
  • Show the diff every time. Do not advance without the user seeing what changed and confirming.
  • User drives the pace. Builder proposes and waits; never barrels through multiple increments unprompted.
  • Quality bar unchanged. Every increment is production-grade — types-first, always-valid domain, boundary .safeParse(), no any, edge cases handled. Pair mode is not an excuse for rough code (that is Forge).
  • Bounded. Max increments (default 12) / user-stop / goal-met / diminishing-returns. On bound, hand the remaining plan back as a normal build plan.
  • Checkpoint-resumable. Persist the increment log so an interrupted session resumes from the last confirmed increment (pair resume).

Verify After Each Increment

Verification is not deferred to the end. After each implemented increment:

  • Show the diff of exactly what changed.
  • Run the increment's verification (the test / type-check / contract check agreed in SETUP or step 1).
  • Report pass/fail; if it introduced a regression or a new edge, surface it as the next increment.
  • Only advance once the increment is green and the user confirms.

The final full 5-axis Impact Scope Check still runs at CLOSE — per-increment verification does not replace it.

Under AUTORUN

Pair mode is INTERACTIVE and cannot run unattended. Under AUTORUN / Nexus AUTORUN:

  • Run SURVEY → PLAN, produce the ordered increment plan + verification paths, and return _STEP_COMPLETE with Next: USER (pair-ready).
  • Do not implement increments without confirmation — that would defeat the mode.

Output

At session end, emit a compact summary:

Pair Programming Summary
- Goal:              [what was being built]
- Implemented:       [increments accepted, with file:line]
- Deferred / skipped:[increments the user set aside]
- Remaining:         [unbuilt increments from the plan]
- Impact Scope:      [5-axis verdict — Ready | Needs Ripple | Blocked]
- Handoff:           [Radar for tests / Guardian for PR / next increment]

Source: SKILL.md on GitHub

No alerts13d5 checks · Risk SAFE
  • Gen Agent Trust Hub13d

    The skill provides a comprehensive environment for production-grade software development, API integration, and AI image-generation pipelines. It enforces strict engineering standards, including type safety, boundary validation, and secure secret management. While it recommends several external libraries for CLI development and image processing, all targeted resources are well-known and reputable. No malicious patterns, obfuscation, or unauthorized data access were detected.

  • Socket13d

    No alerts

  • Snyk13d

    Risk: LOW · No issues

  • Runlayer6mo

    3/8 files flagged

  • ZeroLeaks5mo

    Score: 93/100 · 2 sections analyzed

Signed by skilld at c805268. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 days ago.

Activeupdated 2 weeks ago

README badge

README badge for simota/agent-skills/builder