{
  "markdown": "---\ndocType: guide\nscope: repo\nstatus: current\nauthoritative: true\nowner: skills\nlanguage: en\nwhenToUse: \"When installing, updating, or using the Tiangong AI reusable skills repository.\"\nwhenToUpdate: \"When install commands, target agents, scope behavior, environment variables, or available skill guidance changes.\"\ncheckPaths:\n  - AGENTS.md\n  - .docpact/config.yaml\n  - README.zh-CN.md\n  - .claude-plugin/**\n  - \"*/SKILL.md\"\nlastReviewedAt: 2026-09-07\nlastReviewedCommit: a8e0698e2345e2685d75960bdf972067ecd3b6f2\n---\n\n# Tiangong AI Skills\n\nRepository: https://github.com/tiangong-ai/skills\n\nUse the `skills` CLI from https://github.com/vercel-labs/skills to install, update, and manage these skills.\n\n## Report a problem or suggest a capability\n\nUse the [feedback forms](https://github.com/tiangong-ai/skills/issues/new/choose)\nand [contribution guide](CONTRIBUTING.md). Chinese and English reports are welcome.\nFor CLI runtime problems or uncertain ownership, use the\n[CLI forms](https://github.com/tiangong-ai/cli/issues/new/choose). Auto Research\ncan also prepare the same structured report from its installed reporting reference.\n\n## Atomic data skills\n\nTwenty-one local candidate Skills—AirNow Hourly Observations, EPA EIS Records, Federal Register Documents, NASA\nFIRMS Active Fire, OpenAQ Air Quality, Regulations.gov Comments,\nRegulations.gov Comment Details, Regulations.gov Attachments, USBR Project Records, USBR RISE, USGS Water IV, three Open-Meteo sources, and\nGDELT DOC, Events, GKG, and Mentions, plus Bluesky Cascades, YouTube Video\nSearch, and YouTube Comments—are thin semantic Skills over the Tiangong CLI\nTypeScript 7 data runtime. Each candidate records its stable capability and\noperation contract-major requirements plus any operation feature it actually depends on in\n`references/tiangong-data-requirement.json`; the caller or workspace runtime\nlock chooses the actual CLI build. The agent uses that same resolved CLI for\n`data describe` and `data run`. These Skills contain no second provider\nconnector runtime and no per-Skill package lock.\n\nThis candidate set covers every item in the authoritative EcoCouncil source\nbaseline, `main@ac19289b4876d8a90595a0270721ef3f5ee7ced8`, which contains 21\n`source-fetch` Skills. All 21 have been revalidated against the source\nsemantics, their requirements have been verified against the local CLI\ncandidate, and the unified copy/symlink installation and data-specific gates\npass. The qualifying CLI version and exact manifest/Schema digests live once in\n`scripts/data-skill-migration-provenance.json`; they are release evidence, not\nan installed Skill runtime lock. The repository-wide cold\ngate also passes with a separately scoped upstream filesystem-clock fix; that\nprerequisite is intentionally absent from this data migration branch.\n\nSee `_docs/architecture/atomic-data-capabilities.md` and\n`_docs/runbooks/atomic-data-skill-migration.md` for the ownership boundary,\ncandidate inventory, source audit correction, staged migration order, and\nrelease gates. The audited\nRSS/full-text, Figshare-download, academic-paper, Tiangong/KB, and private-email\ncandidates retain their existing content, artifact, product, research, or\nsecurity boundaries instead of being narrowed into stateless data connectors.\n\n## Install the CLI\n\n```bash\nnpm i skills -g\n```\n\n## Install\n\n- List available skills (no install):\n  ```bash\n  npx skills add https://github.com/tiangong-ai/skills --list\n  ```\n- Install all skills (project scope by default):\n  ```bash\n  npx skills add https://github.com/tiangong-ai/skills\n  ```\n- Install specific skills:\n  ```bash\n  npx skills add https://github.com/tiangong-ai/skills --skill tiangong-auto-research --skill tiangong-kb-sci-search\n  ```\n- Install the Tsinghua graduate thesis LaTeX workflow:\n  ```bash\n  npx skills add https://github.com/tiangong-ai/skills --skill tsinghua-graduate-thesis\n  ```\n- Install the Tiangong KB ingest workflow:\n  ```bash\n  npx skills add https://github.com/tiangong-ai/skills --skill tiangong-kb-ingest\n  ```\n- For a WorkBuddy/CodeBuddy producer, install the thin adapter beside the\n  canonical orchestrator:\n  ```bash\n  npx skills add https://github.com/tiangong-ai/skills --skill tiangong-auto-research --skill tiangong-auto-research-workbuddy\n  ```\n\n## Target agents and scope\n\n- Target specific agents:\n  ```bash\n  npx skills add https://github.com/tiangong-ai/skills -a codex -a claude-code\n  ```\n- Install globally (user scope):\n  ```bash\n  npx skills add https://github.com/tiangong-ai/skills -g\n  ```\n- Scope notes:\n  - Codex is a universal agent: project scope uses `./.agents/skills`, and\n    global scope uses `$HOME/.agents/skills`. `CODEX_HOME` does not change the\n    `skills@1.5.22` global destination.\n  - Claude Code project scope uses `./.claude/skills`; global scope uses\n    `$CLAUDE_CONFIG_DIR/skills` when set, otherwise `$HOME/.claude/skills`.\n  - Other agents have their own directories; inspect the exact path reported by\n    the pinned `skills` CLI rather than deriving `~/<agent>/skills` by analogy.\n\n## Install method\n\n- Interactive installs let you choose:\n  - Symlink (recommended)\n  - Copy\n\n## Update and verify\n\n- List installed skills:\n  ```bash\n  npx skills list\n  ```\n- Check for updates:\n  ```bash\n  npx skills check\n  ```\n- Update all skills:\n  ```bash\n  npx skills update\n  ```\n\n## Environment Variables\n\nEnvironment requirements live with each skill. Before using a skill that calls\nan external service, read that skill's `references/env.md` when present.\n`npx skills add` installs or links skill files; it does not provision language\nruntimes or execute post-install hooks. When a skill provides a locked runtime\nbootstrap and smoke command, run those explicit steps from its own instructions.\n\n## Tiangong KB Ingest Compatibility\n\n`tiangong-kb-ingest` uses the exact reviewed CLI 0.0.48 distribution. After\ninstalling or updating the Skill, run the credential-free compatibility smoke:\n\n```bash\nnpx --yes --package \"@tiangong-ai/cli@0.0.48\" -- tiangong-ai --version\nnpx --yes --package \"@tiangong-ai/cli@0.0.48\" -- tiangong-ai kb --help\n```\n\nThe version command must print `0.0.48`; KB help must list the bulk scan,\nmetadata dry-run, collection list/schema, ingest, and status surfaces. These\nchecks do not call or mutate the KB backend.\n\n## Auto Research External Skill Setup\n\nUse `npx skills` directly for ordinary Skill management. For an Auto Research\nworkspace, prefer the CLI's guarded setup layer: it still uses the exact pinned\n`skills` CLI underneath, while also binding source commits, tree hashes,\ndestinations, license choices, safe credential bindings, and audit state. The\nWizard lets ordinary users enter each selected key with hidden TTY input; named\nenvironment variables and bounded stdin/password-manager input remain explicit\nalternatives. Start with the read-only catalog or the guided Wizard:\n\n```bash\nREVIEWED_BOOTSTRAP_CLI_VERSION=X.Y.Z # replace with one reviewed exact stable release\nnpx --yes --package \"@tiangong-ai/cli@$REVIEWED_BOOTSTRAP_CLI_VERSION\" -- \\\n  tiangong-ai research setup catalog \\\n  --workspace /absolute/path/to/workspace --json\nnpx --yes --package \"@tiangong-ai/cli@$REVIEWED_BOOTSTRAP_CLI_VERSION\" -- \\\n  tiangong-ai research setup \\\n  --workspace /absolute/path/to/workspace\n```\n\nFor repeatable non-interactive provisioning, run `research setup init` first.\nIt creates a no-overwrite schema-v2 `.tiangong-research/setup.yaml` template\nplus `setup.env.example`. The YAML explicitly lists every current catalog Skill,\ncredential, and setting, including disabled optional entries; the env example\nlists every credential variable with an empty placeholder. After the user\nreviews the current catalog, enabled states, licenses, models, pricing, checks,\nand confirmations, bare `research setup` detects only that workspace-local YAML\nand bypasses the Wizard. The optional real `setup.env` must be owner-only;\ndisabled credentials stay empty and a non-empty disabled value is rejected.\nThe removed v1 declaration and all invalid or incomplete declarations fail\nclosed; they never trigger an\ninteractive fallback or parent-directory search. Setup returns success only\nafter every selected dependency, provider live check, Policy compatibility\ncheck, and independent reviewer smoke reaches complete readiness. Use explicit\n`research setup wizard` to choose the interactive path. See\n`tiangong-auto-research/references/setup.md` and\n`tiangong-auto-research/references/env.md`.\n\nThe bootstrap version is an explicit new-workspace choice, never `latest`, a\ntag, or a range. After apply creates `runtime-lock.json`, the installed\norchestrator's bundled resolver runs exactly that locked version for all\nworkspace operations.\n\nFor acquisition-heavy work, the orchestrator uses the CLI's read-only artifact\nsize and role-coverage forecasts, bounded atomic content batches, and explicit\nprepared scientific-review execution. Recovery reuses verified discovery and\ndownloaded artifacts. A compatible locked runtime can revise acquisition in the\nsame project before analysis, explicitly reopening discovery only for new sources;\ndesign changes and later revisions retain the formal successor workflow. Original\nand currently approved task requirements, native check records, independent review\nand portable audit stay linked without extra fixed reviewer rounds. Workflow\nclosure is reported separately from task completion. See\n`tiangong-auto-research/references/execution-assurance.md`; frozen history and\nscientific requirements are never silently rewritten to make a gate pass.\n\nCompatible runtimes also fulfill predeclared pending scientific objects in place,\npreserve original-request provenance, and bind computational acceptance to an\nobserved ordinary native calculation. Complete packet artifacts are available\non demand without an arbitrary total context-length gate. A command report alone\nremains unverified execution; file hashes and read receipts do not certify\nscientific validity. These are CLI-owned contracts, not a second Skill executor.\n\nThe orchestrator gates a research question before setup or any tool call. A\nrequest that presupposes its conclusion or excludes contrary evidence is\npaused with one testable rewrite and resumes only after explicit user\nconfirmation. Controversial topics and directional hypotheses remain allowed\nwhen null results, alternatives, and counterevidence are testable; evidence\nfabrication or concealment is refused.\n\nThe catalog also offers the `tiangong-auto-research` workflow orchestrator,\ndefault-baseline Brave internet evidence, optional Tiangong SCI/document/paper\ncompanions, and optional Anthropic or PPT Master post-closure authoring Skills.\nThe workspace can be any user-selected directory. Every entry is external,\nseparately licensed, pinned, and explicitly confirmed/selected; nothing is\nbundled or installed by a research package. See\n`tiangong-auto-research/references/setup.md` and `external-skills.md`.\nBuilt-in CLI data connectors are different: the native discover packet projects\ntheir current catalog dynamically, and Auto Research invokes the shared\nTypeScript runtime through its Research evidence command. The Skill does not\nduplicate provider adapters or keep a fixed connector list; Research adds only\nbudget, owner-only credential, immutable receipt/ledger, artifact, and review\nbindings around the unchanged core result.\nWhen that result is larger than one Agent context, Auto Research keeps the full\nimmutable evidence and exposes receipt-bound continuation cursors. Exhaustive\nrow review follows those local pages to completion; summary work may stop early\nonly with the presented fraction disclosed. Provider gaps, operation limits,\nand Agent-context projection remain separate signals.\n`tiangong-auto-research-workbuddy` is only a sandboxed-IDE adapter. It routes\nback to the canonical orchestrator and its signed reviewer-bridge reference;\nit does not define a second research workflow.\nFor PPT creation, prefer PPT Master; Anthropic PPTX remains compatible and may\nbe selected alongside it when its workflow fits the task.\n\nFor a top-journal goal, the orchestrator includes a conservative Research\nPolicy template pack for article type, field, journal class, project brief, and\nfour independent final-review roles. The CLI Policy Wizard copies the selected\nMarkdown into the research workspace for human review; it reports when generic\ndefaults remain, requires explicit approval of the exact content hash, and\ninvalidates approval after any edit or expiry. Exact-journal readiness requires\ncurrent official guidance and substantive human customization. These gates can\nproduce a reviewable submission candidate, never a promise of editorial\nacceptance.\n\nBefore discovery, the current native Codex or Claude host must also provide a\nclosed, target-specific scientific design. The CLI validates and freezes the\ndesign. Frozen model implementations and environment locks must first enter the\nworkspace through `research scientific object register`; the Skill never asks\nthe user to hand-copy them into `.tiangong-research`. The CLI requires independent\n`research-design` review before discovery, then real-record `evidence-construct`\nand `pilot-methods` reviews after acquisition and before analysis. After acquisition it also requires exact\ndecomposition records, evidence atoms, a typed-content snapshot, a passing\ninference snapshot, a reproduced analysis, and a mechanically generated\nClaim-Evidence Graph. Publication freeze requires the complete manuscript\nsections and explicit cover/title/checklist/availability/source-data files;\nfour fresh reviews must use the configured agent family that differs from the\nnative producer. It reserves the complete early/final-review and revision\nlifecycle, requires reapproval for every authoritative recovery generation,\nand exports a semantic-chain-verified portable audit directory containing exact\nformal evidence rather than host-local pointers. The CLI remains the\ndeterministic control plane; it never launches a nested producer to invent the\nscience. See\n`tiangong-auto-research/references/publication-policy.md` and\n`tiangong-auto-research/references/scientific-design.md`.\n",
  "bytes": 14210,
  "sha": "d5a68e6fb343ea375755ebe12b966dc2c1c401bef44026b1a6830443d953074f",
  "repo_slug": "tiangong-ai/skills",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/skl_tiangong_ai_skills_email_smtp_send_f6ecf6ee/readme"
}