[EPIC] Canonicalize remaining skills backlog #13

Open
opened 2026-08-22 06:48:39 +00:00 by lost-rob0t · 2 comments
lost-rob0t commented 2026-08-22 06:48:39 +00:00 (Migrated from github.com)

Goal

Turn the remaining known skill drafts/backlog into canonical skills/<id>/ packages without duplicating existing skills or weakening the repository format contract.

This issue is backlog/spec work. Implementation should happen in later RAGE slices, one coherent group at a time.

Canonical format

Every candidate must follow the repository contract before implementation:

  • one durable source under skills/<skill-id>/;
  • SKILL.md YAML frontmatter with name, keyword-only description, and compatibility only when there is a real environment requirement;
  • description: 2–8 comma-separated selection keywords, no prose;
  • body kept small, preferably ~80 lines or less;
  • long examples/reference material move to references/;
  • deterministic helpers move to scripts/;
  • strip personal/private machine assumptions;
  • declare capabilities/dependencies, not author-specific configuration;
  • dedupe by semantic purpose before adding anything;
  • update flake catalog, adapters/index, and CI only when implementation happens.

The upstream Agent Skills spec requires name + description, allows compatibility, metadata, and optional support directories. The repository intentionally keeps description terse while using the body/adapters for richer behavior.

Backlog

GitHub workflow family

  • gh-address-comments
  • gh-close-superseded-prs
  • gh-create-issues
  • gh-fix-ci
  • gh-issue-loop
  • gh-issue-work
  • gh-loop
  • gh-merge-if-green
  • gh-publish-pr
  • gh-triage
  • gh-workflow

Before implementation, build the overlap graph and collapse skills that are only aliases/subsets unless separate activation materially reduces context or authority.

Required semantics from the prior pack:

  • explicit current-request/write gates;
  • requests before backlog;
  • loops continue across the actionable queue;
  • explicit no-progress signatures;
  • UNSTABLE PRs are never classified as merge-ready;
  • no permanent authorization inferred from a loop skill.

Org / task state

  • org-docs — Org-first repository documentation; Markdown only when requested or required.
  • maintain-org-agenda — maintain durable Org agenda/TODO state without creating speculative active work.
  • gpt-todos — durable Org task state for lost-rob0t/gpt-todos, with bidirectional, conflict-aware sync through the user's configured sync/deployment mechanism.

gpt-todos and maintain-org-agenda must be compared for overlap before either is added. Preserve stable IDs/provenance and task-state distinctions such as active, idea, waiting, completed, cancelled, and superseded.

Privileged workflow

  • sudo — portable privileged-operation workflow. Do not bake in the author's desktop/polkit implementation; describe the required authorization capability and preserve the user's configured privilege path.

Explicit dedupe

Do not add rage-loop as a second skill. The canonical rage skill already owns the RAGE workflow. If older rage-loop material contains behavior not represented by rage, reconcile it into the existing skill in a later issue instead of creating a duplicate.

Research gate

Before implementing any backlog item:

  1. inspect the current canonical catalog and nearby semantics;
  2. check current upstream Agent Skills/client conventions;
  3. dedupe against existing repository skills;
  4. prefer the smallest portable operational contract;
  5. only then create an implementation issue/branch.

Acceptance

  • every known backlog candidate is either implemented later, merged into an existing skill, or explicitly rejected as duplicate;
  • no duplicate RAGE skill exists;
  • all implemented skills retain one canonical source and derive client-specific views from adapters;
  • private/user-specific state never enters the repository.
## Goal Turn the remaining known skill drafts/backlog into canonical `skills/<id>/` packages without duplicating existing skills or weakening the repository format contract. This issue is backlog/spec work. Implementation should happen in later RAGE slices, one coherent group at a time. ## Canonical format Every candidate must follow the repository contract before implementation: - one durable source under `skills/<skill-id>/`; - `SKILL.md` YAML frontmatter with `name`, keyword-only `description`, and `compatibility` only when there is a real environment requirement; - `description`: 2–8 comma-separated selection keywords, no prose; - body kept small, preferably ~80 lines or less; - long examples/reference material move to `references/`; - deterministic helpers move to `scripts/`; - strip personal/private machine assumptions; - declare capabilities/dependencies, not author-specific configuration; - dedupe by semantic purpose before adding anything; - update flake catalog, adapters/index, and CI only when implementation happens. The upstream Agent Skills spec requires `name` + `description`, allows `compatibility`, `metadata`, and optional support directories. The repository intentionally keeps `description` terse while using the body/adapters for richer behavior. ## Backlog ### GitHub workflow family - [ ] `gh-address-comments` - [ ] `gh-close-superseded-prs` - [ ] `gh-create-issues` - [ ] `gh-fix-ci` - [ ] `gh-issue-loop` - [ ] `gh-issue-work` - [ ] `gh-loop` - [ ] `gh-merge-if-green` - [ ] `gh-publish-pr` - [ ] `gh-triage` - [ ] `gh-workflow` Before implementation, build the overlap graph and collapse skills that are only aliases/subsets unless separate activation materially reduces context or authority. Required semantics from the prior pack: - explicit current-request/write gates; - requests before backlog; - loops continue across the actionable queue; - explicit no-progress signatures; - `UNSTABLE` PRs are never classified as merge-ready; - no permanent authorization inferred from a loop skill. ### Org / task state - [ ] `org-docs` — Org-first repository documentation; Markdown only when requested or required. - [ ] `maintain-org-agenda` — maintain durable Org agenda/TODO state without creating speculative active work. - [ ] `gpt-todos` — durable Org task state for `lost-rob0t/gpt-todos`, with bidirectional, conflict-aware sync through the user's configured sync/deployment mechanism. `gpt-todos` and `maintain-org-agenda` must be compared for overlap before either is added. Preserve stable IDs/provenance and task-state distinctions such as active, idea, waiting, completed, cancelled, and superseded. ### Privileged workflow - [ ] `sudo` — portable privileged-operation workflow. Do not bake in the author's desktop/polkit implementation; describe the required authorization capability and preserve the user's configured privilege path. ## Explicit dedupe Do **not** add `rage-loop` as a second skill. The canonical `rage` skill already owns the RAGE workflow. If older `rage-loop` material contains behavior not represented by `rage`, reconcile it into the existing skill in a later issue instead of creating a duplicate. ## Research gate Before implementing any backlog item: 1. inspect the current canonical catalog and nearby semantics; 2. check current upstream Agent Skills/client conventions; 3. dedupe against existing repository skills; 4. prefer the smallest portable operational contract; 5. only then create an implementation issue/branch. ## Acceptance - every known backlog candidate is either implemented later, merged into an existing skill, or explicitly rejected as duplicate; - no duplicate RAGE skill exists; - all implemented skills retain one canonical source and derive client-specific views from adapters; - private/user-specific state never enters the repository.
lost-rob0t commented 2026-08-22 06:50:24 +00:00 (Migrated from github.com)

Backlog slices

The recovered backlog is now split into implementation-ready design slices:

  • #15 — GitHub workflow family: 11 candidate skills, overlap/dedupe graph first.
  • #16 — Org/docs/task state: org-docs, maintain-org-agenda, gpt-todos with explicit boundaries.
  • #17 — portable sudo authorization skill.
  • #14 — Agent Zero-native generated format adapter.

rage-loop is intentionally not a backlog item because canonical rage already exists; reconcile any missing semantics into rage rather than minting a clone.

## Backlog slices The recovered backlog is now split into implementation-ready design slices: - #15 — GitHub workflow family: 11 candidate skills, overlap/dedupe graph first. - #16 — Org/docs/task state: `org-docs`, `maintain-org-agenda`, `gpt-todos` with explicit boundaries. - #17 — portable `sudo` authorization skill. - #14 — Agent Zero-native generated format adapter. `rage-loop` is intentionally not a backlog item because canonical `rage` already exists; reconcile any missing semantics into `rage` rather than minting a clone.
Owner

First we should have tea-* skills which are selected by models first.

First we should have tea-* skills which are selected by models first.
Sign in to join this conversation.
No milestone
No project
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
nsaspy/skills#13
No description provided.