@rhize/skill-forge 0.17.1 → 0.20.0

package/LICENSE

@@ -27,6 +27,7 @@ Pro Modules
 - src/refine/store.ts
 - src/refine/patch.ts
 - src/refine/context.ts
+- src/refine/activation.ts
 - src/insights/models.ts
 - src/insights/sources.ts
 - src/insights/store.ts

package/README.md

@@ -2,8 +2,9 @@
 
 **The supply-chain gate for agent skills.**
 
-Status: published — `@rhize/skill-forge@0.17.1` on npm (released 2026-09-05),
-0.x beta (Pro features free until 1.0)
+Source version: `@rhize/skill-forge@0.20.0` (2026-09-14), 0.x beta (Pro features free until 1.0).
+Npm publication is a separate tag-driven step; `npm view @rhize/skill-forge version` reports the
+latest published version.
 
 > This project is unrelated to the [`skillforge`](https://www.npmjs.com/package/skillforge)
 > package on npm, which is a Claude Skills *evaluation* framework. `skill-forge` (this package,
@@ -11,16 +12,17 @@ Status: published — `@rhize/skill-forge@0.17.1` on npm (released 2026-09-05),
 
 ## Why
 
-Agent skills (Claude Code skills, and the equivalent extension formats emerging in Cursor and
-Codex) are now distributed the way npm packages were a decade ago: a public registry, a one-line
-install command, and no vetting step in between. skills.sh alone lists 600k+ skills. A single
-`npx skills@latest add owner/name` drops arbitrary third-party instructions, scripts, and tool
-bindings directly into a working agent's context and file system — with the same trust model as
-copy-pasting a shell script from a stranger.
+Agent skills can include third-party instructions and executable resources. Existing marketplaces
+already provide safeguards: [skills.sh offers partner security audits and installation risk
+information](https://vercel.com/changelog/automated-security-audits-now-available-for-skills-sh).
+Skill Forge adds a local review and maintenance workflow around the skills you actually use:
+provenance, overlap review, drift checks, and recoverable project refinements.
 
-skill-forge does not replace that install command; it wraps it. Every candidate skill is routed
-through **quarantine → profile → safety scan → overlap analysis → report → an explicit
-promote/hold/reject decision**, before it is allowed anywhere near your working skill set.
+Candidates submitted through Skill Forge follow **quarantine → profile → safety scan → overlap
+analysis (when available/configured) → report → an explicit promote/hold/reject decision**.
+This gate applies to the Skill Forge path; it does not prevent direct installs through other tools
+or replace host-managed restrictions. Passing a scan does not prove that a skill is safe.
+See the [product spec](docs/PRODUCT.md) for current competitive positioning and evidence limits.
 
 ## Quickstart
 
@@ -66,7 +68,8 @@ reports every enabled plugin (safety findings, skill count, and — with `--usag
 `unknown` recommendation). Plugin roots are never promotion targets, the safety scan surface
 never shrinks, and nothing here changes an `add`/`scan` verdict. `init` asks once whether to
 enable this; `routine` carries the same flags for cron. See
-[docs/commands/audit.md](docs/commands/audit.md).
+[docs/commands/audit.md](docs/commands/audit.md). For a practical maintenance loop and the limits
+of invocation-based advice, see [Plugin governance](docs/plugin-governance.md).
 
 `<source>` accepts a `skills.sh` `owner/name` slug, a git URL (`https://...`, `git@...`, or
 anything ending in `.git`), or a local filesystem path.
@@ -85,6 +88,7 @@ anything ending in `.git`), or a local filesystem path.
 | `find [query] [options]` | Discover skills via skills.sh and check partner security audits (free) | [docs/commands/find.md](docs/commands/find.md) |
 | `watch [options]` (Pro) | Drift check across every `SOURCES.md` provenance ledger | [docs/commands/watch.md](docs/commands/watch.md) |
 | `ingest [options]` / `queue close <id>` (Pro) | Hand the pending-ingestion queue off to a coding agent for the decide/absorb pass | [docs/commands/ingest.md](docs/commands/ingest.md) |
+| `plugins keep/list/forget` | Record explicit host/project keep rationale without suppressing safety findings | [governance additions](docs/governance-improvements.md) |
 | `refine [options]` (Pro) | Capture, list, review, promote, and roll back project-scope skill overrides | [docs/commands/refine.md](docs/commands/refine.md) |
 | `insight <subcommand>` (Pro) | Turn sources into evidence-backed capability plans and Jira manifests | [docs/commands/insight.md](docs/commands/insight.md) |
 | `routine [options]` (Pro) | One scheduled maintenance pass: audit (plugins + usage aware) + drift + registry, cron-friendly | [docs/commands/routine.md](docs/commands/routine.md) |
@@ -154,8 +158,9 @@ including current caveats, in [docs/configuration.md](docs/configuration.md).
 **Is this related to the `skillforge` npm package?**
 No. [`skillforge`](https://www.npmjs.com/package/skillforge) is a Claude Skills *evaluation*
 framework — it tests whether a skill performs well. `skill-forge` (this package, hyphenated) is a
-supply-chain security gate for skill *installation* — it decides whether a skill is safe and
-non-redundant before it ever runs. Same neighborhood, different problem, name collision only.
+review gate for skill *installation* and ongoing collection maintenance. It reports detected
+risks and potential overlap; it does not certify safety or task effectiveness. The similarly named
+packages are independently maintained.
 
 **Does skill-forge replace `npx skills@latest add`?**
 No, it wraps it. `add` uses the same install mechanisms (skills.sh, git, local copy) but stages
@@ -206,3 +211,26 @@ skill-forge is open-core with a split license (as of v0.2.0):
 
 [LICENSE](LICENSE) is the authoritative map of which files fall under which license.
 Versions up to and including 0.1.0 were published entirely under MIT.
+
+## Governance utility additions (0.19)
+
+Use `plugins keep/list/forget` for explicit host/project retention rationale without
+suppressing safety findings. `audit --plugin-inventory <file>` and the same option
+on `routine` import static host evidence; Codex never joins Claude usage counts.
+`audit --compare <full-audit.json>` highlights changes while preserving all findings.
+`refine activate --skill <capture> --base <dir> --as <unique-name> --host claude|codex`
+materializes a captured patch/extend as a new project skill after a staged safety scan.
+Preview with `--dry-run --json`; activation requires confirmation. Verify actual host
+loading in a fresh task before treating the refinement as consumed. Roll back using
+the returned backup ID with `refine rollback <id> --project --yes`.
+
+See the [governance operating guide](docs/governance-improvements.md) for scope, schemas, examples and limits.
+
+## Advisory instruction diagnostics (0.20)
+
+`audit` reports description, entrypoint and resource measurements separately, with review
+suggestions for broad triggers and resource routing. These suggestions never change safety
+findings, gate decisions or installed skills. Static reports keep actual host loading and
+truncation unknown; rough token estimates are not measured usage. Comparisons show covered
+skills and description/body deltas when both reports contain the new evidence. See
+[the audit command](docs/commands/audit.md) for metric definitions and unavailable-resource handling.