skills: one-line summaries and the delegate-impl rename #79

Merged

vnprc merged 2 commits from agent/skill-summaries into master

2026-10-01 01:31:26 +01:00

allod-agent commented

2026-10-01 01:28:54 +01:00

Member

Every skill in this repo gains a one-line summary: frontmatter field — the line the allod/tools list format shows humans — and delegate-implementation is renamed delegate-impl (directory, frontmatter name, README index, memory.md, and its one reference inside pick-next-issue), fitting the tools-side 16-character name limit. The description: fields are untouched: they stay the agent trigger text, read in full via allod skill <name>.

What changes for a running machine: frontmatter only, plus prose follow-through. README.md now documents the summary/name split (a skill list shows the summary; the description opens with its 100-character lead). memory.md's Skills paragraph says the list shows summaries, and its no-command fallback grep moved from '^description:' to '^summary:' — which is why one body line in skills/forge-groom/SKILL.md that began with summary: (a report-format example, not frontmatter) is rewrapped; without that, the fallback grep would print a seventh, bogus summary.

If it is wrong: the summaries are one line each to edit or revert; the rename is a git mv plus four textual references.

Closes allod/memory#78.

Refs allod/tools#263 — the rendering and the enforced limits (name ≤ 16, summary ≤ 70). Land that first: until a machine rebuilds with it, nothing here breaks anything, and the old list format still reads the untouched descriptions.

Note: the Skills paragraph in memory.md is also touched by the still-open allod/memory#74. Land #74 first (it closes #73) and this rebases onto it, or merge this one and drop #74's paragraph commit — the sentences supersede.

Risk

R0/R1. Frontmatter and prose in a docs repo; no runtime behavior of its own. The only integration point — the fallback grep and the list format — is covered by the rewrapped body line in the same PR.

Validation

.hooks/memory-budget --worktree passes; git grep delegate-implementation returns nothing; the format-A checker against this worktree lists all six skills and exits 0, every line within the limits.
Summaries: 67, 56, 69, 59, 63, 66 characters.
Implemented by a claude-opus worker against a written brief (delegate-implementation); its report is verified line by line by the overseer — checker output, budget hook, rename sweep, and the full diff re-read — and the memory.md/forge-groom follow-through from that report's findings is this PR's second commit.

Every skill in this repo gains a one-line `summary:` frontmatter field — the line the `allod/tools` list format shows humans — and `delegate-implementation` is renamed `delegate-impl` (directory, frontmatter `name`, README index, `memory.md`, and its one reference inside `pick-next-issue`), fitting the tools-side 16-character name limit. The `description:` fields are untouched: they stay the agent trigger text, read in full via `allod skill <name>`. **What changes for a running machine:** frontmatter only, plus prose follow-through. `README.md` now documents the summary/name split (a skill list shows the summary; the description opens with its 100-character lead). `memory.md`'s Skills paragraph says the list shows summaries, and its no-command fallback grep moved from `'^description:'` to `'^summary:'` — which is why one body line in `skills/forge-groom/SKILL.md` that began with `summary:` (a report-format example, not frontmatter) is rewrapped; without that, the fallback grep would print a seventh, bogus summary. **If it is wrong:** the summaries are one line each to edit or revert; the rename is a `git mv` plus four textual references. Closes allod/memory#78. Refs allod/tools#263 — the rendering and the enforced limits (name ≤ 16, summary ≤ 70). Land that first: until a machine rebuilds with it, nothing here breaks anything, and the old list format still reads the untouched descriptions. Note: the Skills paragraph in `memory.md` is also touched by the still-open allod/memory#74. Land #74 first (it closes #73) and this rebases onto it, or merge this one and drop #74's paragraph commit — the sentences supersede. ## Risk R0/R1. Frontmatter and prose in a docs repo; no runtime behavior of its own. The only integration point — the fallback grep and the list format — is covered by the rewrapped body line in the same PR. ## Validation - `.hooks/memory-budget --worktree` passes; `git grep delegate-implementation` returns nothing; the format-A checker against this worktree lists all six skills and exits 0, every line within the limits. - Summaries: 67, 56, 69, 59, 63, 66 characters. - Implemented by a claude-opus worker against a written brief (delegate-implementation); its report is verified line by line by the overseer — checker output, budget hook, rename sweep, and the full diff re-read — and the memory.md/forge-groom follow-through from that report's findings is this PR's second commit.

allod-agent added 2 commits

2026-10-01 01:28:54 +01:00