Skip to main content

volcengine-tls

Use when operating Volcengine TLS (日志服务) through volclog in agent or CI sessions and you need a contract-first workflow for runtime selector resolution, discover/describe/exec routing, App resource or Topic resolution, envelope reading, large-result delivery, and error recovery without guessing commands, fields, or shortcut flows. 当用户提到火山引擎日志服务 TLS、volclog、日志检索/分析/导入/导出、 日志项目或日志主题等操作时调用。

معلومات المصدر

المستودع
volcengine/volcengine-skills
آخر نشاط في المصدر
٢٣ سبتمبر ٢٠٢٦ في ٠٩:١١
لغة SKILL.md المكتشفة
الإنجليزية
النجوم
٢١
التفرعات
٣

خيارات التثبيت

يُحدَّد Prompt الذي يراجع المصدر أولًا بشكل افتراضي. يمكنك التبديل إلى أمر مباشر أو تنزيل نسخة محلية.

مراجعة ملفات المصدر

اقرأ SKILL.md وأي ملفات مرافقة يعرضها SkillsMP قبل أن تقرر التثبيت.

مستكشف الملفات
4 ملفات

عرض SKILL.md

SKILL.md
تعليمات المصدر · معاينة للقراءة فقط
name
volcengine-tls
description
Use when operating Volcengine TLS (日志服务) through volclog in agent or CI sessions and you need a contract-first workflow for runtime selector resolution, discover/describe/exec routing, App resource or Topic resolution, envelope reading, large-result delivery, and error recovery without guessing commands, fields, or shortcut flows. 当用户提到火山引擎日志服务 TLS、volclog、日志检索/分析/导入/导出、 日志项目或日志主题等操作时调用。
license
Apache-2.0
metadata
{"openclaw":{"emoji":"🪵","homepage":"https://github.com/volcengine-tls/ve-tls-cli","requires":{"bins":"[Truncated]"},"install":["[Truncated]"]}}
# Volcengine TLS (volclog) > Adapted from [`volclog-core`](https://github.com/volcengine-tls/ve-tls-cli/tree/b451c506f24573dc0e71220e7ad15339ed7f04e4/skills/volclog-core) in `volcengine-tls/ve-tls-cli` (Apache-2.0). Command examples keep the upstream native `volclog ...` form; run them as `ve volclog ...` (requires `ve >= 1.1.11`) with all later arguments unchanged. ## Overview Use this skill only for agent-only incremental knowledge. Treat this file as the dominant generic operating model for `volclog`, not as an environment-specific bootstrap note and not as an API reference. This file is the behavior summary; references are the authoritative detail. If `volclog tool describe ...` or `volclog workflow describe ...` already answers the question, stop and follow that contract instead of adding duplicate guidance here. ## Operating Stance - Treat `tool describe` or `workflow describe` as the contract truth source. Read `tool describe` or `workflow describe` first. - Treat profile configuration as a runtime-selector problem, not a region-discovery ritual. - Think in runtime selectors: active profile, explicit `--profile <name>`, one-shot `--secrets-file`, `context.secrets_file`, or process-scoped environment credentials. - Use `tool` for published public APIs, `workflow` for CLI-owned orchestration, and `raw` only when method/path is already known. - Keep one-call App, LogApp, Trace, and TraceScore capabilities on `tool`; use `workflow app.resolve-resources` or `app.resolve-topic-ids` only for CLI-owned cross-call expansion. - Prefer structured JSON input such as `--input '{...}'` when `tool exec` or `workflow exec` already supports it. ## Runtime Selector Model - Treat the execution environment as unknown until the available runtime selectors are confirmed. - Do not assume the active or default profile is the right selector for the task. - Run `volclog configure list` only when local profile discovery is actually relevant. - Use `region` only when the contract or task explicitly requires an override. Do not infer it from endpoint or domain. - `--profile`/`context.profile` and `--secrets-file`/`context.secrets_file` are runtime selectors, not business input fields. - Never read, print, export, or persist AK/SK or token values in prompts, memory, request artifacts, or skill content. ## Canonical Agent Loop Discover -> Describe -> Exec -> Read Result 1. Resolve runtime selectors. - decide whether this run is using profile, secrets file, or environment credentials - if local profile discovery matters, run `volclog configure list` - keep the chosen selector explicit in the command or context 2. Discover the callable surface. - `volclog tool list` - `volclog tool list <group>` - `volclog workflow list <group>` 3. Read the contract. - `volclog tool describe <group.action>` - use `volclog tool describe <group.action> --view full` when authoring a nested request or when compact output omits an expected optional field - treat compact view as the discovery summary and full view as the request-authoring contract - `volclog workflow describe <group.command>` 4. Execute with structured input. - `volclog tool exec <group.action> --input '{...}'` - `volclog workflow exec <group.command> --input '{...}'` - `volclog raw --method <METHOD> --path <PATH> --body '{...}'` 5. Read result in this order. - `status` (`"success"` or `"failed"`) - `summary.deliveryMode` - the flat `error` object (when `status` is `"failed"`) - business fields under `data` (when `status` is `"success"`) ## Reference Escalation - Do not read every reference file by default. - Read `tool describe` or `workflow describe` first. - Read [references/routing.md](references/routing.md) only when the surface is still unclear after describe or list. - Read [references/sops.md](references/sops.md) only when the task clearly spans multiple calls and you need stop conditions. - Read [references/best-practices.md](references/best-practices.md) only when runtime semantics, delivery mode, or recovery posture are still unclear after the contract is known. - `contract_cache_hint` and `contract_digest` come from `tool describe` or `workflow describe` output. Read `contract_cache_hint.safe_scope` and `contract_cache_hint.refresh_when`, and reuse cached results only while the same CLI build still reports the same `contract_digest`. - Stop after one reference resolves the current uncertainty; do not keep browsing all references “just in case”. ## Key Rules - Do not guess command names, tool ids, workflow ids, JSON fields, or output shape. - If the action id is still unknown, run `volclog tool list <group>` or `volclog workflow list <group>` before guessing. - Let the operation's `risk`, `recovery`, and `usage_constraints` override the generic recovery map below. - Treat `high-risk-retry` as a warning that retrying is high risk. After an ambiguous result, reconcile only through a get/describe/list action named by the contract; otherwise stop and escalate instead of retrying automatically. - Do not use human shortcut commands as the primary agent flow. - Do not repeat schema details that already exist in `tool describe` or `workflow describe`. - Do not rebuild a valid `tool exec` or `workflow exec` request into flag-heavy shortcut form just because the shell path looks shorter. - Do not pipe `volclog` output into `jq` or `grep` just to rediscover schema or field paths. - Read the full JSON envelope first; use `--jmes-filter` only when you already know the envelope path you want. - Prefer `--dry-run` before any write or destructive change, but only on `raw`, `tool exec`, or `workflow exec`. - Treat `--dry-run` as contract or plan validation, not proof that the business query is correct. - If fields are uniquely named, prefer flat JSON input before manual nesting. ## Error Recovery Quick Map Use this table as the default first response after a failed envelope only after reading the operation's `risk`, `recovery`, and `usage_constraints`. The operation contract takes precedence. Read `references/best-practices.md` only when you need runtime-specific detail beyond the quick action below. | Signal | Next action | | --- | --- | | `error.kind=validation` | Go back to `describe` and fix shape or selector problems first. | | `error.kind=auth` or `401` | Re-check runtime selector, then credentials, then endpoint or region override. | | `403 Forbidden` | Re-check tenant or profile alignment before widening the permission search. | | `error.kind=usage` | Re-run `tool list` or `workflow list`; do not invent aliases. | | `error.kind=unsupported_feature` | Read `error.hint`, then remove the unsupported flag or switch to the named surface; non-LogApp Topic resolution should switch to `app.resolve-resources`. | | `error.kind=incompatible_flags` | Remove one conflicting flag and rerun. | | `error.kind=filesystem` | Fix `--output-dir`, local path, or permissions before retrying. | | `error.kind=decode` | Read `error.hint`, remove the wrong filter or projection if present, then rerun unfiltered or with `--trace-dir`. | | `error.kind=server` | Read `error.code` and `error.requestId`; retry with backoff only when the operation contract permits it and the failure is clearly transient. | | `error.kind=config` | Run `volclog doctor` for host-side configuration checks, then inspect the reported runtime issue. | | `error.kind=unknown` | Read `error.hint` first, then `error.message` or `error.details`, and rerun with `--dry-run` or `--trace-dir` only if more detail is still needed. | | `ResultStatus=incomplete` | Narrow the time window and rerun before trusting counts or emptiness. | `volclog doctor` checks host/runtime configuration such as credentials, endpoint, and local setup. It does not replace `describe`, and it does not validate business query semantics. ## Large Result Handling - Keep surface choice and delivery choice separate. - When a result may be large, provide a writable `--output-dir`. - Let `summary.deliveryMode` tell you what actually happened at runtime. - Read [references/best-practices.md](references/best-practices.md) for exact `outputMode`, `deliveryMode`, file delivery, and `--jmes-filter` semantics instead of duplicating them here. ## Quick Map | Need | Prefer | | --- | --- | | Public API contract already chosen | `volclog tool describe <group.action>` / `volclog tool exec <group.action>` | | CLI workflow already chosen | `volclog workflow describe <group.command>` / `volclog workflow exec <group.command>` | | Need to discover action or workflow ids | `volclog tool list <group>` / `volclog workflow list <group>` | | Need help choosing the surface | read `references/routing.md` | | Need multi-step execution order | read `references/sops.md` | | Resolve an App resource graph | `volclog workflow describe app.resolve-resources` / `volclog workflow exec app.resolve-resources` | | Resolve Topic IDs for a LogApp App | `volclog workflow describe app.resolve-topic-ids` / `volclog workflow exec app.resolve-topic-ids` | | Need runtime/error/recovery semantics | read `references/best-practices.md` | | Use an exact method/path | `volclog raw --method ... --path ...` | | Authenticate a stateless run | host-selected local profile -> one-shot `--secrets-file` or `context.secrets_file` | | Need contract reuse safety | reuse cached results only while the same CLI build still reports the same `contract_digest` |
عرض على GitHub