agentleFS
WardLuWardLu/skills/shadow-skill-publisher/SKILL.md

shadow-skill-publisher

Skill1 starsChanged 2 months agoMITView on GitHub

Validate, package, submit, publish, or track a local Agent Skill through a safe, evidence-backed workflow.

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. About 0 are common, 9 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Follow this sequence exactly:

    line 10
  2. Very specific:

    Start without requiring the user to author profile.json.

    line 14
  3. Very specific:

    If prepare returns needs_browser_observation, use the browser flow in references/browser-execution.md to observe the signed-in creator page …

    line 15
  4. Very specific:

    If prepare returns user_confirmation_required, draft the fields from source and channel evidence, then ask only for the user's final …

    line 16
  5. Very specific:

    Run local checks. When core evidence declares commands, inspect the source-bound execution plan and pass its current digest to prepare …

    line 17
  6. Distinctive:

    Read references/channel-contract.md plus the selected channel reference before describing any remote step.

    line 19
  7. Distinctive:

    Prefer the channel's official API or CLI when the verified contract documents one.

    line 25
  8. Distinctive:

    If no verified API or CLI exists, use a verified browser tool available to the host agent or hand off to any user-controlled browser as …

    line 26
  9. Distinctive:

    Read back the draft or product ID and the exact rendered field values from the remote surface before any submission step.

    line 31
  10. Distinctive:

    If the verified contract has no prior remote-write state, finalize against the adapter-defined local preview ID and the exact preview …

    line 32
  11. Distinctive:

    If automation pauses or contract drift appears, hand off with the exact packet described in references/browser-execution.md.

    line 40
  12. Distinctive:

    Do not ask the user to paste credentials, cookies, tokens, or QR payloads into chat.

    line 41
  13. Distinctive:

    Run the local publisher with:

    line 47

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Shadow Skill Publisher

Use this Skill when validating, packaging, submitting, publishing, or tracking a local Skill. It is agent-neutral and does not require Ego or any browser runtime: local validation, packaging, confirmations, and status tracking run offline. Ego is an optional handoff tool only when a selected channel has no verified API or CLI, the browser surface is part of the documented contract, and the user explicitly authorizes that browser step. Keep the workflow deterministic and evidence-backed. Real platform writes, browser automation, and live channel adapters are limited to documented contracts and handoff instructions.

Follow this sequence exactly:

  1. observe
    • Load the source, evidence profile, channel contract, and channel reference first.
    • Start without requiring the user to author profile.json. When the CLI reports profile.origin=generated, use its source-derived facts. Prompt-only Skills may use their frozen SKILL.md as core source evidence; scripted Skills remain blocked until execution or other verifiable evidence is supplied.
    • If prepare returns needs_browser_observation, use the browser flow in references/browser-execution.md to observe the signed-in creator page and update the private profile on the user's behalf. Do not ask the user to type account aliases or platform fields.
    • If prepare returns user_confirmation_required, draft the fields from source and channel evidence, then ask only for the user's final confirmation. Do not ask them to compose JSON manually.
    • Run local checks. When core evidence declares commands, inspect the source-bound execution plan and pass its current digest to prepare --exec-digest; missing or mismatched authorization remains blocked.
    • Package and verify each channel artifact before building its SubmissionPlan. Preserve successful channel attempts when another channel blocks.
    • Read references/channel-contract.md plus the selected channel reference before describing any remote step.
  2. authorize
    • Show the upload disclosure and the current upload digest verbatim from the generated plan.
    • Record upload authorization only through the authorization workflow that binds the exact current digest.
    • A user saying "同意" or similar is not itself a stored authorization record.
  3. act
    • Prefer the channel's official API or CLI when the verified contract documents one.
    • If no verified API or CLI exists, use a verified browser tool available to the host agent or hand off to any user-controlled browser as documented in references/browser-execution.md. Ego is one optional provider, not a dependency.
    • Open the documented creator page, confirm the visible account, fill only fields from the current plan, upload the exact artifact, read back the rendered values, and stop before the final submit action.
    • Pause instead of guessing whenever the flow reaches login, CAPTCHA, QR scan, platform confirmation, user takeover, or an inactive browser space. A session expiry is login_required, not a request for credentials in chat.
    • Remote writes, prefills, uploads, and registrations stay bound to the exact channel, account_alias, artifact path, and artifact SHA-256 in the current plan.
  4. verify
    • Read back the draft or product ID and the exact rendered field values from the remote surface before any submission step.
    • If the verified contract has no prior remote-write state, finalize against the adapter-defined local preview ID and the exact preview fields before any external write.
    • Unknown remote state is not success. Keep it unknown until read-back proves otherwise.
  5. record
    • Persist the finalized plan after read-back, transition to awaiting_submission_confirmation, and show the new submission digest.
    • Submit or publish only after both current upload and submission authorizations match and the requested action equals the finalized plan's final_action.
    • Record lifecycle progress only from non-empty read-back evidence for the same platform ID and source version; live also requires the public URL.
    • Refresh the human-readable Markdown ledger after every persisted remote transition. Unchanged monitor observations stay write-free and silent.
  6. handoff
    • If automation pauses or contract drift appears, hand off with the exact packet described in references/browser-execution.md.
    • Do not ask the user to paste credentials, cookies, tokens, or QR payloads into chat.

For repeated work, read references/orchestration.md, then use batch to prepare independent source/channel attempts, authorize-batch to bind one acknowledgement to every unchanged digest in that receipt, resume to reconstruct bounded continuation packets, and monitor to ingest read-only observations. Batch authorization never covers a changed or newly added digest. Keep account-bound browser actions serial even when local packaging runs as a batch.

Store reusable commercial defaults, listing assets, and channel artifact exclusions in the private profile described by references/user-config.md. Validate asset digests and verified channel limits before upload. Reuse a known platform_id as an update hint instead of creating a duplicate listing.

Run the local publisher with:

python3 scripts/publisher.py --help

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.