@hasna/skills 0.1.60 → 0.1.61

package/skills/todos-plan/SKILL.md

@@ -1,194 +1,363 @@
 ---
 name: todos-plan
-description: Use when Codewith needs to create, edit, save, verify, execute, route, or sync todos plans for local @hasna/todos work or hosted todos.md/platform-todos work. Triggers include todos plan authoring, local plan Markdown files, plan IDs, task-triggered workers, launch verification checklists, platform plan sync, dry-run plan generation, and updating Codewith workers that must use Todos CLI task IDs as source of truth.
+description: "Create, edit, verify, execute, route, or synchronize Todos plans while preserving provider roles, authenticated authority, full stable IDs, and durable task evidence."
 ---
 
 # Todos Plan
 
 Use the Todos CLI as the source of truth. Markdown plan files are durable
-human-readable artifacts; they do not replace task or plan rows. Never edit
-SQLite, RDS rows, or generated task stores directly.
-
-## Known Launch Plans
-
-- Platform launch plan: `fc27c6c1-7f26-4bea-bbda-dc0731a9d972`
-- Codewith skills plan: `4bc933b6-8b0a-4e38-8807-aa11b034a9af`
-
-When a prompt includes task IDs, inspect and update each task one at a time:
-Commands containing angle-bracket placeholders are illustrative; replace the
-placeholders with real task, plan, org, or file values before running them.
-
-```bash
-todos plans --show <plan-id>
-todos inspect <task-id>
-todos --agent <agent> start <task-id>
-todos comment <task-id> "Progress note"
-todos record-verification <task-id> "<command>" --status passed --summary "<summary>" --agent <agent>
-todos --agent <agent> done <task-id> --files-changed "<paths>" --test-results "<results>" --notes "<notes>"
-```
+human-readable artifacts; they do not replace task or plan rows. Never edit a
+Todos database, hosted row, generated registry, or cached task store directly.
+
+## Required Operation-Authority Gate
+
+Apply this gate before every non-help Todos read or write.
+
+1. Feature-detect the installed surface with metadata-only help:
+
+   ```bash
+   todos --help
+   todos <command> --help
+   ```
+
+   A command or flag absent from installed help is unsupported. Help presence
+   is necessary but not sufficient: the selected authenticated authority must
+   also support the exact invocation. On `REMOTE_COMMAND_UNSUPPORTED` or an
+   equivalent capability error, stop instead of substituting another CLI,
+   bootstrapping a localhost service, extracting internal source, or silently
+   changing the data path.
+2. Record a non-secret provider-role table from package-owned configuration and
+   authenticated readback:
+
+   | Role | Required proof |
+   |---|---|
+   | Product story | `user-hosted` or `Hasna SaaS`. Hasna-operated AWS dogfood is customer-zero of the user-hosted story. |
+   | Client data path | A local SQLite file or the server HTTP API. A client never opens PostgreSQL directly. |
+   | Server operator | The party operating the authenticated Todos server. |
+   | Authenticated authority | Exact API origin, tenant, and principal returned by the installed client. |
+   | Server backend | `sqlite` or `postgresql`, selected by server database configuration. This is storage, not authorization. |
+   | Plan-artifact writer | The package-owned atomic writer and receipt lookup used for Markdown artifacts. |
+
+   These are independent facts. Hasna SaaS is a product story, not a backend.
+   The server backend does not authorize a task operation. Do not derive
+   authority from a location word, provider name, hostname guess, or retired
+   placement enum.
+3. Before production task data access, run the installed authenticated health
+   check:
+
+   ```bash
+   todos health --json
+   ```
+
+   Require a successful authenticated HTTP authority with the intended tenant
+   and no silent fallback. A local SQLite result is valid only for an explicitly
+   selected on-box user-hosted workflow. Ambiguous authority, failed auth,
+   unavailable routing, unsupported commands, or an unexpected backend fails
+   closed. Never fall back from an authenticated production authority to a
+   local file after an error.
+4. Resolve the project through the authenticated authority, capture its full
+   UUID, and prove scoped status:
+
+   ```bash
+   todos --json projects --show <project-ref>
+   todos status --project <full-project-uuid> --json
+   ```
+
+   Require the returned project identity, tenant, and authority to match the
+   intended operation. Never continue from a short ID, stale slug, pathname
+   guess, or cross-project result.
+5. Before an object-specific operation, prove the exact full stable UUID with
+   the corresponding singleton read:
+
+   ```bash
+   todos inspect <full-task-uuid>
+   todos plans --show <full-plan-uuid>
+   todos --json lists --show <full-task-list-uuid>
+   todos template-preview <full-template-uuid>
+   ```
+
+   Use the full UUID returned by the authority for every later read or
+   mutation. If exact identity cannot be proven, stop.
+
+Never print credentials or raw authentication material.
+
+## Production Collection and Completeness Gate
+
+Prefer exact full-ID singleton reads. If a production list, search, history, or
+export is required to resolve or verify an object, use only a package-owned
+producer and reviewed collector that enforce caller- or policy-selected
+positive aggregate caps for calls, items, serialized bytes, and wall time
+before materialization. The collector must own every producer call under the
+remaining budgets, validate complete responses incrementally without uncapped
+buffering, stop at the first excess, and discard the entire result on failure.
+
+Drain every page within those aggregate caps. Reject repeated cursors, invalid
+or partial JSON, trailing unvalidated data, producer failure, timeout, any cap
+hit, scope drift, or a missing or ambiguous terminal signal. A larger cap,
+consumer-side truncation, shell pipeline, generic timeout wrapper, or absence
+from an incomplete result cannot prove completeness. Only an explicit terminal
+marker such as `has_more=false` or equivalent exhaustion metadata can.
+
+## Fresh Production Mutation Gate
+
+Immediately before every production mutation:
+
+1. Re-read the exact full-UUID target and every exact parent object from the
+   same authenticated authority.
+2. Preserve the authority-issued revision, version, ETag, or equivalent
+   compare-and-swap token for the object being changed. Revalidate project,
+   plan, task-list, task, lock, dependency, and approval scope as applicable.
+3. Submit the mutation conditionally against that fresh token. Each mutation,
+   including a comment, verification record, update, approval, completion,
+   deletion, lock change, or dependency change, needs its own fresh pre-read
+   and token. Never reuse a token or treat one successful write as authority
+   for a later write.
+4. Re-read every exact affected UUID from the same authority and verify the
+   intended transition while preserving unrelated fields.
+
+A read followed by an unconditional write is not a drift gate. If installed
+help and the active authority do not expose a version token and conditional
+mutation control for the exact operation, that production mutation is
+non-runnable.
+
+For every production create or append operation, additionally require:
+
+- every exact stable parent UUID advertised by the data model;
+- separate caller-stable operation and step identities plus the
+  authority-documented deterministic C3 key bound to exact immutable
+  semantics, parents, payload, and preconditions; and
+- reconciliation through the same authority by the returned full UUID or an
+  exact idempotency-key lookup.
+
+After a timeout or ambiguous response, reconcile before any retry. One matching
+accepted object may be accepted only after its exact step-bound receipt and
+full readback. Zero accepted effects is not success. A single retry is
+permitted only under C8. Any accepted step enters C9 forward repair or
+receipt-scoped compensation instead of retry. Never issue a blind replacement
+create.
+
+## Named C1-C9 Production Mutation Envelope
+
+Every production create, update, completion, approval, assignment, dependency
+or lock change, delete, comment, verification/evidence record, export, plan
+artifact write, or other state mutation must close one immutable envelope whose
+fields are explicitly named `C1` through `C9`. Existing outcome authorization
+is enough; do not insert a duplicate approval. If the authenticated authority
+or package-owned plan-artifact writer cannot prove a clause, that path is
+non-runnable.
+
+- **C1 Authority, provider roles, full targets, and immediate preimage.**
+  Record the product story, provider-role table, authenticated authority and
+  tenant, server backend, full project UUID, and every full plan, task-list,
+  task, template, export, and artifact target. Immediately before submission,
+  perform bounded complete same-authority reads of every existing target and
+  parent, preserving required full fields and revisions/ETags. For create,
+  require exact bounded terminal authority-issued absence; for an artifact,
+  capture exact absence or the full byte preimage and SHA-256.
+- **C2 Authority-enforced exact-prestate precondition.** Require
+  create-if-absent for row creates/appends and a fresh exact full-target-UUID
+  CAS/`If-Match` for every existing-row mutation. Bind each precondition to the
+  full UUID and complete C1 state or authority-issued revision. Artifact
+  creation must be atomic no-overwrite; replacement must be an atomic expected
+  full-preimage mutation bound to the prior SHA-256 and bytes. Client-side
+  checks, editor saves, shell redirection, and read-then-unconditional-write do
+  not qualify.
+- **C3 Stable distinct step identities and deterministic keys.** Create one
+  caller-stable operation identity and a distinct caller-stable step identity
+  for every ordered manifest entry, including a singleton. Supply the
+  authority-documented deterministic key for each step, bound to operation ID,
+  step ID, ordinal, immutable semantics, all full targets, canonical payload
+  or artifact digest, and exact C1/C2 state. Never invent a key algorithm or
+  reuse identities or keys for a resubmission.
+- **C4 Immutable step-bound receipts.** Require exactly one immutable receipt
+  from the authority or package-owned artifact writer for every ordered step.
+  It must bind operation ID, distinct step ID, ordinal, deterministic key,
+  authority/tenant, full targets, consumed pre-state, canonical payload digest,
+  disposition, accepted effect if any, resulting revision or artifact
+  SHA-256/byte length, and exact receipt identity. Multi-step work also needs an
+  immutable coordinator receipt binding the ordered step receipts and terminal
+  aggregate state.
+- **C5 Exact terminal lookup with hard caps.** Look up the exact receipt,
+  operation, or deterministic key through the same authority under positive
+  finite authority-enforced aggregate caps for calls, items, serialized bytes,
+  and wall time. The artifact writer must provide an equivalent exact terminal
+  receipt lookup. Client-only limits, manual cursor chasing, shell timeouts, or
+  an uncapped filesystem scan do not qualify.
+- **C6 Fail closed.** Do not claim success or execute a dependent step if any
+  cap is missing, zero, unlimited, client-only, or hit; output is partial,
+  incomplete, nonterminal, invalid, or scope-drifted; or lookup returns zero or
+  multiple matches. A terminal zero-effect result is nonacceptance, not
+  success, and may enter only C8.
+- **C7 Exactly-one result, full readback, and duplicate proof.** Require
+  exactly one matching immutable step receipt and exactly one accepted effect
+  for every accepted ordered step, plus the matching coordinator receipt for
+  multi-step work. Perform bounded complete readback of every affected UUID
+  from the same authority and exact full-byte readback of every artifact by
+  path and SHA-256. Verify requested and preserved fields, revisions,
+  identities, ordinals, parent scope, and that each deterministic key maps to
+  the same receipt, target, payload/preimage digest, exact pre-state, and
+  precondition.
+- **C8 Restricted fresh-identity retry.** Allow at most one mutation retry,
+  only after exact C5 lookup proves exactly one immutable terminal
+  nonacceptance receipt for every submitted step and zero accepted effects for
+  the operation. Immediately before resubmission, repeat the full C1 read,
+  capture fresh C2 state, freeze a new ordered manifest, and create a new
+  operation ID, new distinct step IDs, and new deterministic keys. Ambiguous,
+  accepted, zero-match, nonterminal, incomplete, or multiply matched outcomes
+  forbid retry.
+- **C9 Ordered immutable manifest, compensation, and repair.** Before the
+  first write, freeze an ordered immutable manifest, including a one-step
+  manifest for a singleton. Every entry contains its ordinal, operation ID,
+  distinct step ID, dependencies, full targets and parents, exact C1 state, C2
+  preconditions, deterministic key, intended receipt identity, and
+  authority-documented forward-repair or receipt-scoped compensation route.
+  Once a step is submitted, do not rewrite or blind-rerun the manifest.
+  Compensation is permitted only for an exactly identified accepted receipt;
+  every compensating write is a new operation with its own C1-C9 envelope.
+
+Task and plan titles, descriptions, comments, verification commands and
+summaries, completion notes, export contents, plan Markdown, and any other
+private payload must not appear in argv, shell history, diagnostics,
+environment variables, or ad hoc temporary files. Submit bytes only through a
+package-owned protected input channel identified by a non-secret opaque
+descriptor. The descriptor may appear in argv only when it neither reveals nor
+encodes the private value. If the installed package exposes only direct
+positional or flag values for a private field, that production mutation is
+non-runnable.
+
+## Inputs
+
+Resolve these at runtime; none has a built-in default:
+
+- `<full-project-uuid>`: authoritative project UUID.
+- `<full-plan-uuid>`: authoritative plan UUID when the plan exists.
+- `<full-task-uuid>`: authoritative task UUID for the current work item.
+- `<plan-slug>`: caller- or policy-selected stable lowercase slug.
+- `<agent-name>`: inherited agent identity when task state is updated.
+
+Commands containing angle-bracket placeholders are illustrative. Replace every
+placeholder with an authority-proven value before execution.
 
 ## Authoring Workflow
 
-1. Start a native Codewith goal for non-trivial work.
-2. Read the relevant plan with `todos plans --show <plan-id>`.
-3. Inspect the current task before editing anything.
-4. Create or update the local Markdown artifact under
-   `.hasna/todos/plans/<project-id>/<plan-slug>.md`.
-5. Use CLI commands for task/plan state changes:
-   `todos plans --add`, `todos add --plan`, `todos update`,
-   `todos comment`, `todos record-verification`, and `todos done`.
-6. Use the Markdown file for intent, acceptance criteria, dependency notes,
-   rollout notes, and evidence links.
-7. Run validation commands and record them on the task.
-8. Complete only the task whose acceptance criteria are satisfied.
-
-Use direct file edits for plan text. Use Todos CLI for state transitions,
-assignments, comments, dependencies, and verification evidence.
-
-## Local Plan Files
-
-Preferred location:
-
-```text
-.hasna/todos/plans/<project-id>/<plan-slug>.md
-```
-
-Use stable lowercase slugs. Include the Todos plan ID when one exists. If a
-plan is still draft-only, set `plan_id: pending` and create or link the plan
-before workers execute it.
-
-Example:
-
-```markdown
----
-plan_id: 4bc933b6-8b0a-4e38-8807-aa11b034a9af
-project_id: open-skills
-plan_slug: codewith-todos-plan-skills
-status: active
-source: todos-cli
-updated: 2026-06-30
----
-
-# Codewith Todos Plan Skills
-
-## Scope
-- Create a `todos-plan` Codewith skill.
-- Sync it to the active Codewith skills directory.
-
-## Tasks
-- [ ] `4a32ec84` Define authoring workflow.
-- [ ] `36e08375` Document local plan file naming.
-- [ ] `904a4e78` Validate command examples.
+1. Apply the Required Operation-Authority Gate.
+2. Inspect the exact current task and, when present, the exact plan:
+
+   ```bash
+   todos inspect <full-task-uuid>
+   todos plans --show <full-plan-uuid>
+   ```
+
+3. Create or update the relative Markdown artifact through the package-owned
+   protected artifact writer:
+
+   ```text
+   .hasna/todos/plans/<full-project-uuid>/<plan-slug>.md
+   ```
+
+   Pass Markdown bytes by opaque descriptor, not argv or shell redirection.
+   For create, require atomic no-overwrite and a receipt with path, byte length,
+   SHA-256, operation ID, step ID, and deterministic key. For update, supply the
+   exact expected preimage bytes and SHA-256, require atomic replacement, and
+   retain the immutable before/after receipt. Immediately perform exact
+   full-byte readback and hash verification. If the package-owned writer lacks
+   these controls, do not write the artifact.
+4. Use Todos CLI commands for authoritative state. Private values below are
+   represented only by package-owned opaque descriptors:
+
+   ```bash
+   todos add <opaque-task-create-descriptor> --project <full-project-uuid> --plan <full-plan-uuid> <C1-C9-controls>
+   todos update <full-task-uuid> <opaque-update-descriptor> <C1-C9-controls>
+   todos comment <full-task-uuid> <opaque-comment-descriptor> <C1-C9-controls>
+   todos record-verification <full-task-uuid> <opaque-evidence-descriptor> --agent <agent-name> <C1-C9-controls>
+   ```
+
+   The descriptors and `<C1-C9-controls>` are capability requirements, not
+   literal flags. Resolve only package-documented forms from current help and
+   the authenticated authority contract. Run a production form only when every
+   required control exists.
+5. Read back every created or changed object using its returned full UUID.
+6. Record verification as work proceeds. Complete only the task whose
+   acceptance criteria are satisfied:
+
+   ```bash
+   todos --agent <agent-name> done <full-task-uuid> <opaque-completion-descriptor> <C1-C9-controls>
+   ```
+
+   Task completion requires a new exact task read and fresh token after the
+   final preceding mutation.
+
+## Plan Artifact Requirements
+
+The relative Markdown artifact must include:
+
+- full Todos project, plan, and task UUIDs where applicable;
+- scope and out-of-scope boundaries;
+- acceptance criteria and dependencies;
+- rollout or recovery notes when relevant;
+- verification commands and evidence references; and
+- residual risks.
+
+Draft-only artifacts must say that no authoritative plan row exists. Do not
+invent placeholder UUIDs or allow workers to execute against a draft identity.
 
-## Status Transitions
-- `pending`: task is not started.
-- `in_progress`: one agent owns it and has a lock.
-- `blocked`: a named prerequisite is missing.
-- `completed`: verification is recorded and accepted.
-
-## Verification
-- `python3 .../quick_validate.py skills/todos-plan`
-- `todos plans --show 4bc933b6-8b0a-4e38-8807-aa11b034a9af`
-```
-
-## Hosted Sync
-
-Hosted sync is explicit. Confirm tenant, auth, and billing before writing.
-
-```bash
-platform-todos --json auth status
-platform-todos --json auth whoami
-platform-todos --json billing status
-platform-todos docs catalog --surface api --json
-platform-todos docs catalog --surface mcp --json
-```
+## Worker Routing
 
-Preview first:
+Task-triggered workers receive:
 
-```bash
-platform-todos plans generate \
-  --org <organization-id> \
-  --objective "Create project tasks, dependencies, and verification gates" \
-  --dry-run true \
-  --approval-before-create true
-```
+- full task, plan, and project UUIDs;
+- repository root plus repo-relative allowed and out-of-scope paths;
+- Todos CLI source-of-truth instruction;
+- validation and evidence requirements; and
+- adversarial verification requirement.
 
-Create only after approval:
+Use isolated work scopes. Do not route new repo-mutating work through prompt
+paste. Native Codewith goals and loops are Codewith mechanisms; OpenLoops
+workflows and runs are separate and must be named and recorded as OpenLoops
+only when that system is actually used.
 
-```bash
-platform-todos plans generate \
-  --org <organization-id> \
-  --objective "Create project tasks, dependencies, and verification gates" \
-  --dry-run false \
-  --approval-before-create true
-```
+## Destructive Plan Operations
 
-For local imports, prefer a dry run and a conflict strategy:
+Plan completion or deletion requires exact scope already authorized by the
+caller and the Fresh Production Mutation Gate:
 
 ```bash
-platform-todos import local-sqlite <manifest.json> \
-  --org <organization-id> \
-  --dry-run true \
-  --conflict-strategy skip \
-  --idempotency-key <stable-key>
+todos plans --show <full-plan-uuid>
+todos plans --complete <full-plan-uuid> <C1-C9-controls>
+todos plans --delete <full-plan-uuid> <C1-C9-controls>
 ```
 
-Record evidence on the task: command, status, org/project ID, plan ID, and
-artifact path. Do not record API keys, raw tokens, secret values, or private
-payloads.
-
-## Worker Routing
-
-Task-triggered launch workers must receive:
-
-- task ID and plan ID
-- repo path and allowed paths
-- source-of-truth instruction: use Todos CLI, not messages or tmux panes
-- native Codewith goal requirement
-- validation gates and evidence expectations
-- explicit out-of-scope paths
-- requirement for adversarial verification before completion
-
-Use isolated work scopes. Do not route new repo-mutating work by pasting prompts
-into existing tmux panes. If the routing automation is missing, create a task
-for the missing automation instead of using a hidden fallback.
-
-Prompt shape:
-
-```text
-Task: <task-id>
-Plan: <plan-id>
-Repo: <absolute-path>
-Allowed paths: <paths>
-Out of scope: <paths>
-Use Todos CLI as source of truth.
-Start a native Codewith goal.
-Record verification and changed files on the task before done.
-```
-
-## Launch Checklist
-
-For todos.md launch tasks, verify the relevant surfaces and record evidence:
+These forms are syntax references only. If the client cannot submit the exact
+operation conditionally against the freshly read plan version, do not run it
+in production. Re-read the exact plan after completion. After deletion require
+an authoritative exact-ID tombstone or not-found result distinguishable from
+auth, routing, transport, and incomplete collection failure.
 
-- Auth: `platform-todos auth login`, `auth status`, `auth whoami`, API key
-  creation/revocation/rotation.
-- Billing: `platform-todos billing status`, `usage`, `checkout`, `portal`,
-  Stripe webhook lifecycle, quota errors.
-- Plan CRUD: local `todos plans --add/show`, hosted `platform-todos plans
-  templates/create/generate/refine`, dry-run before create.
-- Hosted runs: list/create/show/logs/artifacts/cancel plus controls, pause,
-  resume, emergency stop, usage evidence, and sandbox policy.
-- Audit export: redacted export command, hash/manifest evidence, no secrets.
-- Storage: signed upload/download/delete/export behavior and quota gates.
-- Landing and docs: desktop/mobile screenshot review, docs links, billing entry,
-  no unsupported dashboard or OAuth promises.
+Do not add a duplicate approval gate when the caller already authorized the
+exact outcome. If the exact plan identity or destructive scope is new, stop.
 
 ## Done Criteria
 
-- The plan file is under `.hasna/todos/plans/<project-id>/`.
-- Todos CLI reflects the task/plan state.
-- Command examples were run or clearly marked illustrative.
-- Verification is recorded with `todos record-verification`.
-- Changed files and residual risks are noted on the task.
-- For substantial work, an adversarial verifier or explicit adversarial
-  self-review is reconciled before marking the task done.
+- The Required Operation-Authority Gate passed.
+- The provider-role table, authenticated authority, and server backend were
+  recorded as independent facts.
+- Every required collection was producer-bounded and explicitly terminal
+  without hitting an aggregate cap.
+- All commands and flags exist in installed help and on the active authority.
+- The plan artifact is relative and linked to authority-proven full UUIDs.
+- Every production mutation has a complete immutable named C1-C9 envelope.
+- Every artifact create/update used atomic no-overwrite or exact-preimage
+  semantics, immutable receipts, and full byte/hash readback.
+- Every existing-object mutation used a fresh same-authority CAS token and was
+  read back from the same authority.
+- Every create or append used exact parent UUIDs, separate operation/step
+  identities, the authority-documented deterministic C3 key, and bounded
+  terminal reconciliation before success was claimed.
+- Verification, changed files, and residual risks are recorded.
+- Adversarial verification is reconciled before completion.
+
+## Related
+
+- [[todos-plans]] — plan, template, and task-list CLI operations.
+- [[todos-progress]] — comments, locking, dependencies, and history.
+- [[todos-filter]] — bounded task listing, filtering, and export views.

package/skills/video-highlight-pack/SKILL.md

@@ -1,6 +1,7 @@
 ---
 name: video-highlight-pack
-description: Generate premium video highlight packages with clip plans, captions, thumbnails, chapter markers, social copy, edit decisions, and manifest metadata.
+description: Generate video highlight packages with clip plans, captions, thumbnails, chapter markers, social copy, edit decisions, and manifest metadata.
+kind: instruction
 ---
 
 # Video Highlight Pack
@@ -9,28 +10,25 @@ Generate a practical highlight package from a long video, recording, webinar, de
 
 ## Requirements
 
-- Authenticate with `skills auth login`.
-- This premium skill runs through the hosted Skills runtime; local installs expose only metadata and instructions.
+None. This is an instruction skill: it is prose an agent follows, so it needs no
+credentials, no network access, and no local runtime.
 
 ## Usage
 
-```bash
-skills run video-highlight-pack --source ./webinar.mp4 --title "AI billing launch webinar" --platforms "youtube-shorts,linkedin"
-skills run video-highlight-pack --source ./transcript.txt --duration-minutes 58 --aspect-ratio 9:16
-```
+Ask an agent for the deliverables below and give it the inputs. Reading this file
+IS the invocation; `skills run` refuses instruction skills on purpose.
 
-## Options
+## Inputs
 
-| Option | Description | Default |
+| Input | Description | Default |
 |--------|-------------|---------|
-| `--source <path-or-text>` | Video file, transcript text file, or transcript text. | required |
-| `--title <text>` | Recording title. | Video Highlight Pack |
-| `--platforms <list>` | Comma-separated platforms for export planning. | youtube-shorts,instagram,tiktok,linkedin |
-| `--duration-minutes <n>` | Approximate source runtime for timestamp spacing. | 45 |
-| `--aspect-ratio <ratio>` | Primary edit aspect ratio. | 9:16 |
-| `--output <dir>` | Output directory for direct local execution. Hosted runs use the run export directory. | run export dir |
+| `source` | Video file, transcript text file, or transcript text. | required |
+| `title` | Recording title. | Video Highlight Pack |
+| `platforms` | Comma-separated platforms for export planning. | youtube-shorts,instagram,tiktok,linkedin |
+| `duration-minutes` | Approximate source runtime for timestamp spacing. | 45 |
+| `aspect-ratio` | Primary edit aspect ratio. | 9:16 |
 
-## Outputs
+## Deliverables
 
 - `highlight-plan.md`
 - `clips.csv`
@@ -41,4 +39,16 @@ skills run video-highlight-pack --source ./transcript.txt --duration-minutes 58
 - `edit-decision-list.json`
 - `manifest.json`
 
-After submitting a hosted run, poll with `skills runs status <run-id>` and download outputs with `skills exports download <run-id>`.
+## Method
+
+1. Work from a transcript. If given a media file, transcribe it first (the
+   `audio-transcript-pack` skill, or a local `ffmpeg` + transcription step); this
+   skill plans from text.
+2. Select highlights by self-contained value: a clip that needs context to make
+   sense is not a highlight.
+3. Give each clip precise in/out timecodes and a reason for selection.
+4. Write captions to the platform's line-length limits and keep them readable at
+   speed.
+5. Set chapter markers at genuine topic boundaries, not at fixed intervals.
+6. Write per-platform social copy that stands alone without the video.
+7. Deliver the edit decision list so an editor can cut without re-watching.

package/skills/video-highlight-pack/package.json

@@ -5,7 +5,6 @@
   "description": "Hosted premium video highlight package generator",
   "type": "module",
   "skills": {
-    "runtime": "hosted",
-    "source": "remote"
+    "kind": "instruction"
   }
 }

package/dist/cli/onboarding.d.ts

@@ -1,13 +0,0 @@
-import type { Command } from "commander";
-import { type SkillsConfig } from "../lib/config.js";
-export interface FirstRunOnboardingInput {
-    argv: string[];
-    commandName?: string;
-    config: SkillsConfig;
-    isInteractive: boolean;
-    testMode?: boolean;
-}
-export declare function getRootCommandName(actionCommand: Command): string;
-export declare function shouldShowFirstRunOnboarding(input: FirstRunOnboardingInput): boolean;
-export declare function getFirstRunOnboardingMessage(): string;
-export declare function maybePrintFirstRunOnboarding(actionCommand: Command, argv: string[], isInteractive: boolean): void;

package/dist/lib/hosted-availability.d.ts

@@ -1,13 +0,0 @@
-import type { SkillAvailabilityMetadata } from "./registry-types.js";
-export interface HostedRunUnavailable {
-    ok: false;
-    status: 503;
-    code: "HOSTED_PROVIDER_UNAVAILABLE";
-    message: string;
-    details: string[];
-}
-export type HostedRunAvailability = {
-    ok: true;
-} | HostedRunUnavailable;
-export declare function getHostedRunAvailability(slug: string): HostedRunAvailability;
-export declare function getHostedAvailabilityMetadata(slug: string): SkillAvailabilityMetadata;