name	repo-worker-edit-verify
description	Execute repo-local code edits and run deterministic verification.
metadata	{"short-description":"Edit code with a tight repo-worker verify loop","tags":["coding","editing","repo-worker","testing"]}

Instructions

Apply the minimal edits needed for the current step while preserving unrelated behavior.
Treat existing tests as constraints and do not weaken or rewrite them without an explicit behavior-level reason.
When changing validation, invariants, or error behavior, update all neighboring code paths that implement the same rule so callers observe consistent semantics.
Replace assertion-based public input checks with explicit runtime errors when the task is about user-facing validation semantics; keep assertions only when they are intentionally internal invariants. Do not add new public-input validation with assert.
If a search or command result is truncated, rerun a narrower command or use exact file reads before editing.
Run targeted checks first, then broader checks only when the task requires them.
Add focused regression coverage when a missing behavior guard is the safest way to lock in the fix.
Record command-output summaries with explicit pass, mixed, blocked, failed, or not-run outcomes and note unresolved risks.
Keep shell commands simple and Host-Mount-safe: avoid &&, shell globs, and brace expansion; use separate tool calls when you need multiple reads or checks.
Prefer repo-local runners or interpreters when available, such as .venv/bin/python, venv/bin/python, python -m pytest, tox, or nox; if the environment is clearly unavailable, report environment_blocked instead of continuing blind retries.
Do not install dependencies, sync environments, or invoke approval-requiring bootstrap commands solely to get verification running. If the available environment cannot execute the targeted checks directly, stop and report environment_blocked.

Más de este repositorio

mismo repositorio

swebench

realmorrisliu/alan

Run and inspect SWE-bench benchmark workflows for alan's coding line. Use this when: - The user wants to run SWE-bench Lite or curated SWE-bench subsets - The user wants to prepare benchmark workspaces or materialize suite manifests - The user wants to score benchmark predictions with the official harness - The user wants benchmark operations to remain separate from the repo-coding product contract

2026-07-169

repo-coding

realmorrisliu/alan

Delegate bounded repo-scoped coding work to a fresh repo worker. Use this when: - alan has already selected the target repo or directory - The task needs inspect -> plan -> edit -> verify -> deliver inside that bound scope - The work should run in a focused child runtime instead of the parent steward

2026-07-139

memory

realmorrisliu/alan

Manage persistent Memory Store files across Agent Processes. Use this when: - Continuing work started by an earlier Agent Process - The user references prior decisions, constraints, or history - Stable, user-confirmed information should be retained - Durable handoff or topic notes need inspection or maintenance

2026-07-139

alan-shell-control

realmorrisliu/alan

Inspect and operate the native alan shell control surface. Use this when: - The user asks about spaces, tabs, surfaces, panes, focus, or splits - alan needs to decide which pane or surface should receive an action - alan needs to create a space, split a pane, focus a pane, or send text - The task depends on understanding the outer terminal app rather than only understanding Alan Process state This skill assumes the local `alan shell` CLI namespace backed by the shell's IPC/socket API. If that CLI is unavailable, report that clearly and fall back to other available context.

2026-07-119

plan

realmorrisliu/alan

Create and manage structured execution plans for complex multi-step tasks. Use this when: - Task requires more than 3 distinct steps - Work is expected to span multiple context windows - There are logical dependencies between steps - You need to track incremental progress - User asks for something complex that benefits from a plan Do NOT use for simple single-step tasks or quick questions.

2026-07-119

skill-name

realmorrisliu/alan

Describe what this delegated skill does and when it should be used.

2026-07-109

name	repo-worker-edit-verify
description	Execute repo-local code edits and run deterministic verification.
metadata	{"short-description":"Edit code with a tight repo-worker verify loop","tags":["coding","editing","repo-worker","testing"]}

Instructions

Apply the minimal edits needed for the current step while preserving unrelated behavior.
Treat existing tests as constraints and do not weaken or rewrite them without an explicit behavior-level reason.
When changing validation, invariants, or error behavior, update all neighboring code paths that implement the same rule so callers observe consistent semantics.
Replace assertion-based public input checks with explicit runtime errors when the task is about user-facing validation semantics; keep assertions only when they are intentionally internal invariants. Do not add new public-input validation with assert.
If a search or command result is truncated, rerun a narrower command or use exact file reads before editing.
Run targeted checks first, then broader checks only when the task requires them.
Add focused regression coverage when a missing behavior guard is the safest way to lock in the fix.
Record command-output summaries with explicit pass, mixed, blocked, failed, or not-run outcomes and note unresolved risks.
Keep shell commands simple and Host-Mount-safe: avoid &&, shell globs, and brace expansion; use separate tool calls when you need multiple reads or checks.
Prefer repo-local runners or interpreters when available, such as .venv/bin/python, venv/bin/python, python -m pytest, tox, or nox; if the environment is clearly unavailable, report environment_blocked instead of continuing blind retries.
Do not install dependencies, sync environments, or invoke approval-requiring bootstrap commands solely to get verification running. If the available environment cannot execute the targeted checks directly, stop and report environment_blocked.