Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/amd-aim/inference-skill/inferencex-optimizenpx skills add AMD-AIM/inference-skill --skill inferencex-optimizegit clone --depth 1 https://github.com/AMD-AIM/inference-skillWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00042 | $0.01380 |
| Opus 5 | $0.00021 | $0.00690 |
| Sonnet 5 | $0.00008 | $0.00276 |
| Haiku 4.5 | $0.00004 | $0.00138 |
Grade A, and why
inferencex-optimize scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 107 lines — stays where its author put it; the contents beside it link to each section on GitHub.
InferenceX Optimize
Default user experience
- Treat a bare model/config key as enough to start. Do not require the user to spell out a full command.
- If the user says
use inferencex-optimize skill for <model-or-config-key>, start guided setup immediately. - Do not dump raw parameter names in the first reply; translate to a short setup conversation.
- Prefer the native
questiontool for multiple-choice prompts when the runtime provides it. - If the runtime does not provide a question tool, ask concise numbered choices in normal chat.
- Ask questions in grouped batches, not as a drip-feed of one question at a time.
- Keep explicit progress updates so the user always knows current stage and next step.
- Inform user about GEAK availability during setup when running in optimize or optimize-only mode.
First-turn latency rule
- Do not read any other file before the first visible reply unless the model/config name is ambiguous.
- The first visible reply should happen immediately:
- send one short kickoff status update
- ask the first grouped setup form
- Do not read
INTAKE.md,RUNTIME.md, orEXAMPLES.mdbefore the first grouped form. - Only after the user answers Round 1 should you read deeper reference files.
Guided setup flow
- Resolve the target config key from the user's model/config name.
- Start with one short high-level question round. The first question groups should be exactly:
Run planOutputGPUs
- After Round 1 answers, read
INTAKE.mdand follow its deeper intake flow. - Read
RUNTIME.mdonly when you are about to do discovery or execution bootstrap. - Ask high-level setup questions first, then do lightweight config discovery, then ask filter-specific questions.
- Do not ask TP / sequence length / concurrency until discovery has produced concrete options.
- For smoke runs, offer a fast-path
Filterschoice:Use recommended smoke defaultsReview each filterUse full discovered sweep
- Summarize the final plan and get a clear go/no-go from the user with a
Confirmquestion before executing. - Only after confirmation, read the needed phase docs and start execution.
What ships with it
28 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- EXAMPLES.md 3.3 KB
- INSTALL.md 2.2 KB
- INTAKE.md 11 KB
- LICENSE 1.0 KB
- phases/00-env-setup.md 6.2 KB
- phases/01-config-parse.md 4.4 KB
- phases/02-benchmark.md 8.0 KB
- phases/03-benchmark-analyze.md 5.8 KB
- phases/04-profile.md 23 KB
- phases/05-profile-analyze.md 43 KB
- phases/06-problem-generate.md 13 KB
- phases/07-kernel-optimize.md 21 KB
- phases/08-integration.md 25 KB
- phases/09-report-generate.md 7.8 KB
- resources/TraceLens-internal.tar.gz 4902 KB
- RUNTIME.md 6.7 KB
- scripts/analyze_fusion_inferencex.py 11 KB runs code
- scripts/classify_kernel.py 12 KB runs code
- scripts/generate_problems_inferencex.py 60 KB runs code
- scripts/generate_sglang_plugin.py 10 KB runs code
- scripts/generate_vllm_plugin.py 15 KB runs code
- scripts/kernel_finalize.py 1.9 KB runs code
- scripts/kernel_test_runner.py 9.2 KB runs code
- scripts/select_gpus.py 7.3 KB runs code
- scripts/trace_analyzer.py 63 KB runs code
- templates/agent-config.md 4.9 KB
- tests/e2e_optimize_test.py 42 KB runs code
- tests/E2E_TEST.md 8.5 KB
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 107 lines · 42 tokens per session scan A d4cc05c784dd
inferencex-optimize is a skill published in the GitHub repository AMD-AIM/inference-skill (5 stars, last pushed 3mo ago), licensed MIT. It adds 42 tokens to every session and 1,380 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
instrument-data-to-allotrope
Convert laboratory instrument output files (PDF, CSV, Excel, TXT) to Allotrope Simple Model (ASM) JSON format or flattened 2D CSV. Use this skill when scientists need to standardize instrument data for LIMS systems, data lakes, or downstream analysis. Supports auto-detection of instrument types. Outputs include full…
exploratory-data-analysis
Perform bounded, local exploratory analysis of explicitly supported scientific files. Use for redacted CSV/TSV/JSON profiles; optional NumPy, HDF5, FASTA/FASTQ, and basic image metadata inspection; missingness/leakage audits; outlier and transformation sensitivity; and rigorous EDA report scaffolds. Other domain…
mapping-to-snomed
Maps clinical concept spans extracted by OpenMed to SNOMED CT concepts through a USER-SUPPLIED terminology server (the user's own Ontoserver, Snowstorm, or UMLS/UTS), never a bundled vocabulary. Use when the user wants to code findings, disorders, procedures, body structures, or substances to SNOMED CT, run an ECL…
auditing-subgroup-fairness
Audit an OpenMed NER or de-identification model for performance disparities across demographic subgroups (sex, age band, race/ethnicity when available) using openmed.eval.fairnessreport. Use when the user wants per-subgroup recall and leakage, wants to check whether de-identification under-protects a group, wants to…
overleaf-sync
Two-way sync between a local paper directory and an Overleaf project, so ARIS audit/edit workflows stay on the local copy while collaborators edit in the Overleaf web UI. Use when user says "同步 overleaf", "overleaf sync", "推送到 overleaf", "connect overleaf", "Overleaf 桥接", "pull overleaf", "push overleaf", or wants to…
mixed-precision
Use FP16/BF16 mixed precision to accelerate training and reduce memory. Use when optimizing GPU performance.