Ingest a skill

Ingest is the protocol path for receiving a portable skill: inspect trust, validate integrity, then load or dry-run. Ask your agent — same order on any conforming host.

Prompt your agent

I have a file at ./file.skill. Inspect TrustView (digests, seals) without executing.
Validate integrity, then dry-run. Summarize what it does and any trust warnings.
Do not execute for real unless I explicitly ask.

Continuity resume:

Load ./handoff.skill as continuity context. Summarize intent, scrubbed journey,
open gaps, and pinned knowledge. Resume the work; do not mint a fake release.

Needs the CLI installed first — run skill --version, and if that fails, npm i -g skillerr.

What your agent will do

Safe ingest order:

skill inspect ./file.skill --trust
skill validate ./file.skill
skill verify-trust ./file.skill
skill run ./file.skill                 # dry-run by default

Step	What you learn
inspect --trust	Manifest, digests, seal, TrustView — no execution
validate	Structure + hash integrity
verify-trust	Seal / issuer checks
run (dry-run)	Planned steps without side effects

Never feed untrusted package bodies into a model before TrustView.

Continuity: load in another agent

skill load ./file.skill

The agent receives intent, redacted journey, open questions, scrubbed knowledge, completeness gaps, and typed inputs — not raw transcripts or credentials.

Execute (opt-in)

Execute mode refuses unsigned, open, public-dev HMAC (development), and self_reported seals unless the host explicitly allows untrusted run (--allow-untrusted). Use that only when you knowingly accept the risk.

For public-dev reference mints during inspection: skill verify-trust ./file.skill --allow-development-issuer.

Runtime behavior (summary)

Lifecycle: LoadAndVerify → TrustView → NegotiateCapabilities → ResolveInputs → Consent → Execute → Verify → EmitSkillRun.

Modes: inspect, explain, dry_run, execute, resume.

Deny-by-default: undeclared network, filesystem outside declared roots, undeclared secret slots, missing consent.

Unpack (inspection only)

skill unpack ./file.skill — prefer digests from inspect / validate over eyeballing unpacked files as a trust decision.

Trust reminders

• Unsigned = untrusted
• Public-dev HMAC = development, forgeable — not production identity
• Host / model fields under a seal are still claims, especially for local LLMs

Details: Trust and security