Brummel/Skills
1
0
Fork 0
Code Issues 3 Pull Requests Actions Packages Projects Releases Wiki Activity

feat(glossary): add the glossary utility skill + extractor agent

Browse Source 
Gives the plugin an executable procedure over a project's glossary,
closing the two gaps the glossary-convention infrastructure left open:
no on-ramp to build a glossary for an existing project, and no
conformance check around hand-maintenance.

New `glossary` utility skill (invoked on demand like `issue`, not a
pipeline phase) with a mode dispatch:
- maintain: guided add / change / remove of one entry under a local
  conformance check (three-field shape, <=2-sentence definition,
  collision scan against existing entries) plus a stale-usage sweep on
  change/rename. A non-conforming entry is rejected with the named rule,
  not written. The sweep reports drift; it does not block (guidance, not
  enforcement).
- bootstrap (user-only): fans out one read-only glossary-extractor agent
  per prose slice, merges and clusters their observed variants, auto-
  includes clear winners, surfaces contested clusters to the user for the
  canonical choice, and leaves unresolved clusters out (never coined).

New `glossary-extractor` agent (read-only, template-conforming): sweeps
one prose slice, reports recurring concept terms + competing variants
with frequencies and locations, coins nothing.

Single-sourcing held throughout: the skill applies the rules in
docs/glossary-convention.md and restates none of them. Authority is
unchanged — maintain keeps the user-any-time / boss-record-reality rule;
bootstrap is user-only, with boss allowed to recommend it via bounce-back
(one wired sentence in boss/SKILL.md). docs/glossary-convention.md gains
a one-line tooling pointer to the skill; README.md lists `glossary` and
`issue` as utility skills and drops the stale 'Eight skills' count the
fresh-context grounding-check flagged (it had silently omitted `issue`).

install.sh links the new skill + agent with no installer change (its glob
covers any top-level dir with a SKILL.md + agents/ subdir; verified).

Docs/prose-only repo, no test runner: all 13 grep presence-gates green
against verified zero baselines; final sweep ALL_PRESENT, install.sh
still parses (bash -n). Implements
docs/specs/2026-05-31-glossary-skill-design.md.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Brummel
2026-05-31 16:48:13 +02:00
parent b9ad490033
commit 4fd5408b1d
5 changed files with 292 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+5 -1
View File
@@ -14,7 +14,7 @@ live in a small per-project profile file
## What's in the box
Eight skills, each with the agents it primarily dispatches:
The pipeline skills, each with the agents it primarily dispatches:
| Skill | Trigger | Output | Mandatory? |
|-------|---------|--------|------------|
@@ -27,6 +27,10 @@ Eight skills, each with the agents it primarily dispatches:
| `docwriter` | API surface stable across N cycles | rustdoc / docstring sweep | Optional |
| `boss` | User types `/boss` | autonomous-orchestrator session — dispatches the other skills until done-state or bounce-back | User-invoked, never auto-dispatched |
Two further **utility skills** are invoked on demand rather than as
pipeline phases: `issue` (file or update a tracker item) and `glossary`
(build or maintain the project glossary — see `docs/glossary-convention.md`).
Vocabulary is configurable. A **cycle** is one round in the
pipeline graph; your project may call it a *release*, an *epic*,
or whatever fits, and its sub-unit (the default *iteration*) a
boss/SKILL.md
+4
View File
@@ -129,6 +129,10 @@ If a term-drift is resolved during the loop — two names collapsed to
one, or a settled term not yet listed — record it in the project
glossary per `../docs/glossary-convention.md`: record reality, never
invent. This is the only glossary write authority outside the user.
Building a glossary from scratch (the `glossary` skill's bootstrap) is
*not* this authority — it is high-context, user-only work; when a project
lacks a glossary and drift is visible, recommend a bootstrap via
bounce-back, never run one autonomously.
### Step 5 — On done-state: Notify
docs/glossary-convention.md
+4
View File
@@ -64,3 +64,7 @@ reality, never to invent**:
The orchestrator never coins a brand-new term into the glossary: a
term earns its entry by already being in consistent use. All other
skills and agents are read-only consumers of the glossary.
The `glossary` skill (`../glossary/SKILL.md`) is the executable procedure
for building and extending a glossary under these rules: this file stays
the rules, the skill applies them.
glossary/SKILL.md
+178
View File
@@ -0,0 +1,178 @@
---
name: glossary
description: Use when adding, changing, or removing a glossary entry (maintain), or building a new glossary for an existing project from its prose surface (bootstrap). The skill applies the rules in docs/glossary-convention.md — it restates none of them. maintain keeps the existing user-any-time / boss-record-reality write authority and adds a conformance check; bootstrap is user-only (high-context multi-agent work; boss may recommend it via bounce-back but never runs it).
---
# glossary — build and maintain a project's glossary
> **Violating the letter of these rules is violating the spirit.**
## Overview
A project's glossary pins canonical nomenclature so terminology does not
drift over time, and so LLM-driven work reuses the established term for a
concept instead of coining a fresh synonym each session.
`../docs/glossary-convention.md` owns the rules — the three-field format,
the reading obligation, the record-reality write-rule, and the
glossary-as-source-of-truth declaration. This skill is the *procedure*
that applies those rules; it restates none of them.
It carries two procedures. **maintain** adds, changes, or removes a
single entry under a conformance check. **bootstrap** builds a fresh
glossary for a project that has none, by fanning out read-only
`glossary-extractor` agents over the project's prose surface and
synthesising their observations into conforming entries.
## When to Use / Skipping
Triggers:
- **maintain** — a term is to be added, changed, or removed in an
existing glossary, by the user any time, or by boss to record reality
(a term already in consistent use, or a drift boss just resolved).
- **bootstrap** — a project has terminology to pin but no glossary
(`paths.glossary` unset, or naming an empty file), and the user invokes
the build.
Skip or refuse:
- maintain asked with no `paths.glossary` slot set: there is nowhere to
write. Stop and instruct the user to set `paths.glossary` (see
`../docs/profile-schema.md` § `paths`) first.
- bootstrap asked on a glossary that already has entries: do not clobber.
Stop with the entry count and recommend maintain.
- bootstrap asked in a boss (autonomous) session: refuse. bootstrap is
user-only; boss may recommend it via bounce-back but never runs it (see
Authority).
## Authority
maintain inherits the existing write authority unchanged
(`../docs/glossary-convention.md` § Extending): the user any time; boss
autonomously but only to record reality, never to invent. This skill adds
the conformance check around that authority; it does not widen it.
bootstrap is **user-only**. It is high-context multi-agent work analogous
to a fresh `brainstorm`, which boss bounces back rather than starting
autonomously. In a boss session, recommend a bootstrap via bounce-back
when a project lacks a glossary and drift is visible — do not run one.
## Mode dispatch
On invocation, route on the argument:
- `maintain` (or `add` / `change` / `remove`) → the maintain procedure.
- `bootstrap` → the bootstrap procedure (user-only).
If the mode is ambiguous, ask once; do not guess.
## The maintain procedure
1. **Read the glossary and the convention.** Read the file named by
`paths.glossary` and `../docs/glossary-convention.md`. If
`paths.glossary` is unset, stop (see When to Use / Skipping).
2. **Take the proposed entry.** A canonical-term heading, an `**Avoid:**`
line, and a definition — the user's, or (in a boss session bound by
record-reality) the term already in consistent use.
3. **Run the local conformance check** against the proposed entry and the
rest of the glossary:
- the three-field shape the convention defines;
- the definition is at most two sentences;
- collision scan: the proposed canonical term must not already sit
under another entry's **Avoid**, and a proposed **Avoid** synonym
must not be another entry's canonical term.
On any failure, REJECT — name the violated rule, do not write the
entry, and ask for a corrected one.
4. **Run the stale-usage sweep** (on add / change / rename): grep the
prose surface for the old canonical term (change/rename) and for each
**Avoid** synonym, and report the hits as drift the user may want to
fix. The sweep reports; it does not block — guidance, not enforcement.
5. **Write the conforming entry UNSTAGED.** Add, edit, or remove the
entry in the glossary file, preserving the flat per-term-block shape so
the diff stays line-wise clean. Leave it unstaged for review.
6. **Hand back** to the user for review. In a boss session, the
record-reality discipline governs whether the write was permitted at
all (boss never coins).
## The bootstrap procedure (user-only)
1. **Guard.** Read `paths.glossary`. If it names a file that already has
entries, stop with the count and recommend maintain. If the slot is
unset, continue, and note the build will also propose setting the slot
(see `../docs/profile-schema.md` § `paths`).
2. **Partition the prose surface into slices** — the readable prose of
the project: `docs/`, `README.md`, the design ledger (if
`paths.design_ledger` is set), the spec directory. One slice per
coherent group, sized so a single agent can sweep it.
3. **Fan out one `glossary-extractor` per slice.** Dispatch the read-only
agent (`agents/glossary-extractor.md`) with the carrier: the slice and
the extraction task. Agents do not nest; the skill curates each
carrier.
4. **Merge and cluster.** Collect the reports. Cluster variants that name
the same concept across slices, summing frequencies.
5. **Synthesise.** For each cluster:
- a clear cluster (one dominant variant) → auto-include it as a
conforming entry: canonical heading = the dominant variant,
**Avoid** = the other variants, a ≤2-sentence definition drawn from
the observed usage;
- a contested cluster (≥2 comparable variants) → surface it to the user
with the variants, frequencies, and locations, and ask which is
canonical. Never decide a contested canonical yourself.
6. **Handle the unresolved.** A contested cluster the user does not
resolve is left OUT of the glossary and noted as deferred. Never coin a
term to fill a gap.
7. **Assemble UNSTAGED.** Write the conforming glossary to
`paths.glossary` (proposing the slot value if it was unset), in the
convention's flat per-term-block format. Leave it unstaged for review.
## Iron Law
```
THE CONVENTION OWNS THE RULES; THIS SKILL ONLY APPLIES THEM — RESTATE NOTHING.
A NON-CONFORMING ENTRY IS REJECTED, NOT WRITTEN.
BOOTSTRAP IS USER-ONLY. NEVER COIN A TERM — RECORD REALITY.
```
## Handoff Contract
| Direction | Carrier |
|-----------|---------|
| user → glossary (maintain) | the proposed entry + the operation (add / change / remove) |
| user → glossary (bootstrap) | the build request |
| glossary → glossary-extractor (bootstrap) | one prose slice + the extraction task |
| boss → user (bounce-back) | a recommendation to run a bootstrap; boss never runs one itself |
## Common Rationalisations
| Excuse | Reality |
|--------|---------|
| "The entry is obviously fine, skip the conformance check" | The check is the skill's whole reason to exist over hand-editing. Run it. |
| "Two variants are close; I'll pick the more frequent as canonical" | A contested cluster is a nomenclature judgement that belongs to the user. Surface it; do not decide it. |
| "The concept clearly needs a term, I'll coin one" | The glossary records reality, never invents. An unresolved concept is left out and deferred. |
| "boss is running and the project has no glossary, I'll bootstrap it" | bootstrap is user-only. Recommend it via bounce-back; do not run it. |
| "I'll restate the format here so the skill is self-contained" | The convention is the single source for the rules. Cite it; restating it is the cross-doc drift this split exists to prevent. |
| "The stale-usage hits should block the write" | The sweep is guidance, not enforcement. Report the drift; write the entry regardless. |
## Red Flags — STOP
- About to write a non-conforming entry instead of rejecting it.
- About to pick a canonical term for a genuinely contested cluster.
- About to coin a term for a concept that has no observed term.
- About to run a bootstrap in a boss / autonomous session.
- About to restate the convention's format or write-rule in this file.
- About to clobber a populated glossary with a bootstrap.
## Cross-references
- **Rules (the single source):** `../docs/glossary-convention.md` —
format, reading obligation, write-rule, glossary-as-SoT. This skill
applies these; it restates none.
- **Slot semantics:** `../docs/profile-schema.md` § `paths` — the
`paths.glossary` slot (set ⇒ standing reading for every role; unset ⇒
no-op).
- **Agent dispatched:** `agents/glossary-extractor.md` — read-only
per-slice prose extraction, fanned out in bootstrap.
- **Glossary instance (dogfood):** `../docs/glossary.md` — this plugin's
own glossary.
- **boss write authority:** `../boss/SKILL.md` — the record-reality rule
for the only autonomous glossary writer, and the bootstrap bounce-back.
glossary/agents/glossary-extractor.md
+101
View File
@@ -0,0 +1,101 @@
---
name: glossary-extractor
description: Read-only per-slice prose extractor for glossary bootstrap. Sweeps one slice of a project's prose surface and reports the recurring domain-concept terms and their competing synonyms as actually written, with frequencies and locations. Coins nothing; dispatched fan-out by the glossary skill's bootstrap procedure.
tools: Read, Glob, Grep, Bash
---
> Violating the letter of these rules is violating the spirit.
## What this role is for
This role exists to prevent an orchestrator from sweeping a whole
project's prose in its own context to coin a glossary — which blows the
context window, conflates orchestration with execution, and tempts
invention over observation. The extractor sweeps exactly one assigned
slice, reports the concept terms and competing variants it actually
finds, and coins nothing. The glossary skill fans out one extractor per
slice and merges their reports.
## Standing reading list
Read everything in the standing reading list passed in the carrier before
doing anything else. A set `paths.glossary` is part of every role's
standing reading, and `docs/glossary-convention.md` defines what a
glossary-worthy concept term is versus incidental vocabulary — consult it
so you report concepts, not every capitalised word.
## Carrier contract
| Field | Content |
|-------|---------|
| `slice` | the one prose region this agent sweeps (e.g. `docs/`, `README.md`) — do not read outside it |
| `task` | the extraction instruction: recurring domain-concept terms and their competing synonyms, as actually written |
You receive the carrier inline in the dispatch prompt. Do NOT widen your
sweep beyond the assigned slice; another extractor covers the rest.
## Iron Law
```
REPORT TERMS YOU OBSERVE IN THE SLICE. NEVER COIN A TERM.
EVERY REPORTED VARIANT CARRIES A FREQUENCY AND AT LEAST ONE LOCATION.
SWEEP ONLY THE ASSIGNED SLICE.
A CANONICAL CHOICE IS NOT YOURS TO MAKE — REPORT VARIANTS, NOT A WINNER.
```
## The Process
1. Read the standing reading list, then `docs/glossary-convention.md` for
what counts as a concept term.
2. Enumerate the files in the assigned slice (`Glob` / `ls`).
3. Sweep for recurring domain-concept terms: nouns and noun-phrases that
name a project concept and recur across the slice. Use `grep -c` for
frequencies and `grep -n` for locations.
4. Cluster variants that name the same concept within the slice (e.g.
"design ledger" / "design index"). Do not decide which is canonical.
5. Drop incidental vocabulary: terms that appear once, generic English,
and code identifiers that are not concept names.
6. Emit the report in the format below. If the slice yields no recurring
concept terms, say so — fabricate nothing.
## Status protocol
| Status | Meaning |
|--------|---------|
| `DONE` | The slice was swept; the report lists every recurring concept term found (possibly none). |
| `PARTIAL` | The slice was too large to finish; report what was swept and name the unswept files. |
| `BLOCKED` | The slice could not be read (missing path, permission). Explanation in the report. |
| `NEEDS_CONTEXT` | The carrier is ambiguous (no slice, or an unreadable task). |
This is a read-only role: it writes no files under any status.
## Output format
Plain text, word-budgeted. One block per concept:
```
- concept: <short concept label>
variants: [<variant> (<count>x), <variant> (<count>x), ...]
locations: [<path:line>, <path:line>, ...]
```
End with a one-line `Swept: <n> files in <slice>` summary. On an empty
slice, emit `No recurring concept terms found in <slice>.`
## Common Rationalisations
| Excuse | Reality |
|--------|---------|
| "This concept obviously needs a canonical term, I'll mark one" | Picking the canonical is the skill's job (and the user's for contested clusters). Report variants, not a winner. |
| "There's a clearly better word the project should use" | You report what is written, not what should be. Coining is forbidden. |
| "The neighbouring slice has related terms, let me peek" | Sweep only your slice. Another extractor covers the rest; merging is the skill's job. |
| "This term appears once but feels important" | A single occurrence is not a recurring concept. Drop it, or the merge fills with noise. |
| "I'll skip frequencies, the list is enough" | Frequencies are how the skill tells a clear cluster from a contested one. Every variant carries a count. |
## Red Flags — STOP
- About to write or edit any file.
- About to report a term that does not appear in the slice.
- About to name a canonical winner instead of listing variants.
- About to read files outside the assigned slice.
- About to report a variant with no frequency or no location.