Get Available Resources
Windows Runtime Commands
On native Windows, use the managed Windows runner and the native runtime command target. Set $runtime to the installed runtime root. Multi-agent installs usually use %LOCALAPPDATA%\ai-agents-skills\runtime. Then run:
$runtime = if ($env:AAS_RUNTIME_ROOT) { $env:AAS_RUNTIME_ROOT } else { "$env:LOCALAPPDATA\ai-agents-skills\runtime" }
& "$runtime\run_skill.ps1" "skills/get-available-resources/detect_resources.py" <args>
POSIX examples below use run_skill.sh and .sh command targets; use the Windows command target above on native Windows.
Use this skill before local work that may be expensive, memory-sensitive, or
parallelized, such as document conversion batches, graph enumeration, SageMath
runs, OCR, local parsing, or large file rearrangement.
Workflow
- Decide whether the task is heavy enough to justify a preflight. Skip this
skill for trivial commands.
- Prefer an existing local resource checker when the installed agent provides
one. Otherwise inspect resources with portable system commands or Python.
- Record the result in a small planning note or
.agent_resources.json in the
current workspace when the task will continue for multiple steps.
- Use the result to choose batch size, parallelism, memory strategy, and
whether to route the task to SageMath, WSL, remote compute, or a smaller
local run.
Minimum Checks
- CPU count and rough CPU model. Report the cores this process may use, not the
machine's: a CPU affinity mask or a cgroup quota bounds a run below the online
core count and shows up in neither
/proc/cpuinfo nor psutil.cpu_count.
- Available memory, bounded the same way by the cgroup limit when one applies.
/proc/meminfo reports a machine that a container will be OOM-killed well below.
- Free disk space in the working directory. A probe that fails reports the failure
and the run continues; it does not cost the checks that succeeded.
- GPU or accelerator availability only when relevant and detectable.
- Whether the workload should be split, sampled first, or routed elsewhere.
Output Shape
For a visible preflight, report:
- resources inspected
- detected limits
- recommended execution strategy
- confidence and any missing probes
Guardrails
- Do not spend more time on resource detection than the task warrants.
- Do not assume GPU or SageMath availability without checking.
- On Windows, consider WSL-backed tools separately from native Windows tools.
- Treat remote compute credentials and provider configuration as external; do
not inspect or print secrets.
Recommended templates
When this skill is involved, consider these workflow templates (install via
the workflow-templates artifact profile, or --with-deps to pull backing skills):
autonomous-research-loop-runbook -- Bounded autonomous research-loop runbook with four stop conditions, single-path solving, mandatory cross-agent verification, fresh-agent backtracking, and five-lane broker-routed heavy-compute offload with per-lane safety gates.engineering-delivery-loop-runbook -- Bounded build-and-deliver loop runbook: single-path implementation with seen-to-fail proof, cross-agent diff verification, behavior-preserving cleanup, and five-lane broker-routed heavy-compute offload with per-lane safety gates.
1---2name: get-available-resources3description: Use at the start of computationally intensive local tasks to detect CPU, memory, disk, and optional accelerator availability before planning execution.4---56# Get Available Resources789## Windows Runtime Commands1011On native Windows, use the managed Windows runner and the native runtime command target. Set `$runtime` to the installed runtime root. Multi-agent installs usually use `%LOCALAPPDATA%\ai-agents-skills\runtime`. Then run:1213```powershell14$runtime = if ($env:AAS_RUNTIME_ROOT) { $env:AAS_RUNTIME_ROOT } else { "$env:LOCALAPPDATA\ai-agents-skills\runtime" }15& "$runtime\run_skill.ps1" "skills/get-available-resources/detect_resources.py" <args>16```1718POSIX examples below use `run_skill.sh` and `.sh` command targets; use the Windows command target above on native Windows.1920Use this skill before local work that may be expensive, memory-sensitive, or21parallelized, such as document conversion batches, graph enumeration, SageMath22runs, OCR, local parsing, or large file rearrangement.2324## Workflow25261. Decide whether the task is heavy enough to justify a preflight. Skip this27 skill for trivial commands.282. Prefer an existing local resource checker when the installed agent provides29 one. Otherwise inspect resources with portable system commands or Python.303. Record the result in a small planning note or `.agent_resources.json` in the31 current workspace when the task will continue for multiple steps.324. Use the result to choose batch size, parallelism, memory strategy, and33 whether to route the task to SageMath, WSL, remote compute, or a smaller34 local run.3536## Minimum Checks3738- CPU count and rough CPU model. Report the cores this process may use, not the39 machine's: a CPU affinity mask or a cgroup quota bounds a run below the online40 core count and shows up in neither `/proc/cpuinfo` nor `psutil.cpu_count`.41- Available memory, bounded the same way by the cgroup limit when one applies.42 `/proc/meminfo` reports a machine that a container will be OOM-killed well below.43- Free disk space in the working directory. A probe that fails reports the failure44 and the run continues; it does not cost the checks that succeeded.45- GPU or accelerator availability only when relevant and detectable.46- Whether the workload should be split, sampled first, or routed elsewhere.4748## Output Shape4950For a visible preflight, report:5152- resources inspected53- detected limits54- recommended execution strategy55- confidence and any missing probes5657## Guardrails5859- Do not spend more time on resource detection than the task warrants.60- Do not assume GPU or SageMath availability without checking.61- On Windows, consider WSL-backed tools separately from native Windows tools.62- Treat remote compute credentials and provider configuration as external; do63 not inspect or print secrets.6465## Recommended templates6667When this skill is involved, consider these workflow templates (install via68the `workflow-templates` artifact profile, or `--with-deps` to pull backing skills):6970- `autonomous-research-loop-runbook` -- Bounded autonomous research-loop runbook with four stop conditions, single-path solving, mandatory cross-agent verification, fresh-agent backtracking, and five-lane broker-routed heavy-compute offload with per-lane safety gates.71- `engineering-delivery-loop-runbook` -- Bounded build-and-deliver loop runbook: single-path implementation with seen-to-fail proof, cross-agent diff verification, behavior-preserving cleanup, and five-lane broker-routed heavy-compute offload with per-lane safety gates.
hoanganhduc