nw-ab-critique-dimensi…
Review dimensions for validating agent quality - template compliance, safety, testing, and priority validation
Queues a deferred self-update of nwave-ai. Writes a PendingUpdateFlag that the SessionStart hook replays on the next Claude Code launch, so the current session is not interrupted. Falls back to manual instructions when the package manager cannot be detected.
$ npx -y skills add nWave-ai/nWave --skill nw-update --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/nw-updateContext preview
The summary Claude sees to decide when to auto-load this skill.
Queues a deferred self-update of nwave-ai. Writes a PendingUpdateFlag that the SessionStart hook replays on the next Claude Code launch, so the current session is not interrupted. Falls back to manual instructions when the package manager cannot be detected.
name: nw-update description: "Queues a deferred self-update of nwave-ai. Writes a PendingUpdateFlag that the SessionStart hook replays on the next Claude Code launch, so the current session is not interrupted. Falls back to manual instructions when the package manager cannot be detected." user-invocable: true argument-hint: '[target-version] - Optional explicit version (e.g. 1.4.2). Defaults to the latest version discovered by the periodic update check.'
**Wave**: CROSS_WAVE | **Agent**: Main Instance (self) | **Command**: `/nw-update`
`/nw-update` records a `PendingUpdateFlag` on disk. The actual upgrade is performed by the SessionStart hook on the next Claude Code launch, so the user can keep working in the current session and apply the new version whenever they restart.
You (the main Claude instance) run this command directly — no subagent delegation. Do NOT attempt to upgrade nwave-ai in the current session.
1. If the user passed an argument (`$ARGUMENTS`), treat it as the target version after stripping a leading `v`. 2. Otherwise read the latest version discovered by the periodic update check from the global config (`update_check.latest_available`):
python3 -c "
import sys, os
sys.path.insert(0, os.path.expanduser('~/.claude/lib/python'))
from des.adapters.driven.config.des_config import DESConfig
print(DESConfig().update_check_latest_available or '')"3. If no argument AND no discovered version (empty output): ask the user for the target version explicitly. Do not guess.
Invoke the detector and resolve the PM binary absolute path in a single Bash call.
This runs under whatever `python3` the shell provides, which is NOT the interpreter nwave-ai was installed under — so `sys.executable` is the wrong anchor for detection. Use `resolve_nwave_pm`, which prefers the value the installer recorded at install time (`install.package_manager` in the global config) and only falls back to live path detection:
python3 -c "
import sys, os
sys.path.insert(0, os.path.expanduser('~/.claude/lib/python'))
import shutil
from pathlib import Path
from des.adapters.driven.config.des_config import DESConfig
from des.adapters.driven.package_managers.package_manager_detector import (
resolve_nwave_pm,
)
pm = resolve_nwave_pm(DESConfig().installed_package_manager, Path(sys.executable))
binary = shutil.which(pm) if pm in ('pipx', 'uv') else ''
print(f'{pm}|{binary or \"\"}')"Parse the output as `pm|binary_abspath`.
If `pm == "unknown"` OR `pm == "pip"` OR `binary_abspath` is empty:
(`detect_pm` returns `pip` only when forced via `NWAVE_INSTALLER=pip`; pip installs are not auto-updatable through the deferred-upgrade flow, so they take the manual path alongside `unknown`.)
1. Do NOT call `PendingUpdateService.request_update`. 2. Print the manual fallback and stop:
nwave-ai was not installed via an auto-updatable package manager.
Please upgrade manually:
uv tool install nwave-ai@latest && nwave-ai install
— or —
pipx upgrade nwave-ai && nwave-ai install
— or, if you installed with pip —
pip install -U nwave-ai && nwave-ai install
Then restart Claude Code.Call the driving port `PendingUpdateService.request_update`:
Reuse the `pm` and `binary` already resolved in Step 2 — do NOT re-detect from `sys.executable` here (that interpreter is not the install anchor). Construct `DESConfig()` directly; there is no `DESConfig.load()`.
python3 -c "
import sys, os
sys.path.insert(0, os.path.expanduser('~/.claude/lib/python'))
from des.application.pending_update_service import PendingUpdateService
from des.adapters.driven.config.des_config import DESConfig
pm = '${PM}'
binary = '${PM_BINARY}'
target = '${TARGET_VERSION}'
config = DESConfig()
# PM adapter is only used by apply(); request_update() does not invoke it.
svc = PendingUpdateService(config=config, pm=None) # type: ignore[arg-type]
svc.request_update(pm=pm, pm_binary_abspath=binary, target_version=target)
print(f'queued:{pm}:{target}')"Substitute `${PM}` and `${PM_BINARY}` with the Step 2 values, and `${TARGET_VERSION}` with the Step 1 value (no leading `v`).
On success, print:
Update to vX.Y.Z queued. It will be applied the next time you restart Claude Code. Package manager: <pipx|uv> (<binary_abspath>)
| Condition | Response | |-----------|----------| | No target version provided and none discovered | Ask the user for an explicit version | | PM detection returns `unknown` | Print manual fallback (Step 3), do not write flag | | `shutil.which(pm)` returns empty | Same as `unknown` — print fallback | | `request_update` raises | Surface the exception, do not claim success |
This command is synchronous and fast (< 1s). Do not create a task list; print progress inline.
/nw-update
Reads `update_check.latest_available = "1.4.2"`, detects `pipx` at `/home/alex/.local/bin/pipx`, queues flag, prints: `Update to v1.4.2 queued. It will be applied the next time you restart Claude Code.`
/nw-update 1.5.0
Detects `uv` at `/home/alex/.cargo/bin/uv`, queues flag for `1.5.0`.
/nw-update 1.4.2
Detector returns `unknown` (e.g. pip venv install). Prints manual fallback instructions and exits without writing the flag.
AI agents that guide you from idea to working code, with human judgment at every gate. nWave runs inside Claude Code. It breaks feature delivery into seven waves (discover, diverge, discuss, design, devops, distill, deliver).
Repo: nWave-ai/nWave
Review dimensions for validating agent quality - template compliance, safety, testing, and priority validation
Review dimensions for validating agent quality - template compliance, safety, testing, and priority validation
Review dimensions for acceptance test quality - happy path bias, GWT compliance, business language purity, coverage completeness, walking skeleton…
Detailed 5-phase workflow for creating agents - from requirements analysis through validation and iterative refinement
5-layer testing approach for agent validation including adversarial testing, security validation, and prompt injection resistance
Architectural style selection decision matrices, trade-off analysis, structural enforcement rules, and combination patterns. Load when choosing or evaluating…