@event4u/agent-config 9.8.0 → 9.9.0

package/docs/benchmark.pinned.yml

@@ -1,31 +0,0 @@
-# Pin manifest for the curated composite `docs/benchmark.md`
-# (road-to-flow-learnings Phase 3).
-#
-# Each section renders INDEPENDENTLY from ITS pinned report via
-# `src/scripts/render_benchmark_composite.ts` — never from the single
-# latest report (the composite's in-file curation rationale stands).
-# Curated prose lives OUTSIDE the `<!-- pinned:<id> -->` markers and is
-# never touched by the renderer. Pinned reports are operator-local
-# (untracked); the renderer skips cleanly where they are absent.
-sections:
-  - id: honesty
-    mode: honesty-labels
-    report: internal/bench/reports/ab-v2/2026-06-15T03-52-35Z-ab-v2-paired.json
-  - id: weak-stats
-    mode: stats-body
-    report: internal/bench/reports/ab-v2/2026-06-15T03-52-35Z-ab-v2-paired.json
-  - id: cost-factor-table
-    mode: arm-cost-table
-    report: internal/bench/reports/ab-v2/2026-07-07T05-35-14Z-ab-v2-paired.json
-    arms:
-      - id: vanilla
-        label: none
-      - id: rules-balanced
-        label: kernel + tier 1 (shipped `balanced` profile)
-      - id: rules-kernel-dc
-        label: kernel (9 rules) + `downstream-changes`
-      - id: package
-        label: full plugin
-  - id: strong-table
-    mode: compact-table
-    report: internal/bench/reports/ab-v2/2026-07-05T07-00-31Z-ab-v2-paired.json

package/docs/benchmarks.md

@@ -1,85 +0,0 @@
----
-stability: beta
-keep-beta-until: 2026-08-14
----
-
-# Benchmark cadence
-
-> **Status:** active · **Owner:** `step-16-telegraph-substance.md` Phase 1 ·
-> **Sources:** [`benchmark-corpus-spec.md`](contracts/benchmark-corpus-spec.md) ·
-> [`benchmark-report-schema.md`](contracts/benchmark-report-schema.md)
-
-Where the package's benchmark runs live, when they run, and what counts as
-a publishable report. Mirrors the the external runtime `docs/benchmarks/runs/<ISO>.json`
-discipline (upstream `5b71c7a`).
-
-## Corpora
-
-| Corpus | Path | Purpose |
-|---|---|---|
-| `dev` | `tests/eval/corpus-dev.yaml` | router / engine selection |
-| `telegraph` | `internal/bench/corpora/telegraph/prompts.yaml` | condensation dialect (`vs_raw` + `vs_terse`) |
-| `rtk` | `internal/bench/corpora/rtk/commands.yaml` | rtk CLI-output filtering savings (Phase 2 of `road-to-readable-value-dashboard.md`) |
-| `value` | _derived_ | aggregated dashboard — no own corpus, reads from the others |
-
-## Reports — naming and trail
-
-- **Canonical pointer:** `internal/bench/reports/<corpus>-v<N>.{json,md}` — always
-  reflects the latest published run for that corpus version.
-- **Timestamped trail:** `internal/bench/reports/<ISO-Zulu>-<corpus>-v<N>.{json,md}`
-  — every committed run keeps an immutable history copy alongside.
-
-Both are produced in one `scripts/bench_run.py` invocation; do not commit
-one without the other.
-
-## Cadence
-
-| Trigger | Required corpus | Required artefact |
-|---|---|---|
-| Pre-release bake (any `vX.Y.0`) | `dev` + `telegraph` | both reports refreshed |
-| Edit to `src/rules/telegraph-speak.md` | `telegraph` | report refreshed in same PR |
-| Edit to `scripts/bench_run.py` `--telegraph` arm | `telegraph` | report refreshed in same PR |
-| Edit to `internal/bench/corpora/telegraph/prompts.yaml` | `telegraph` | report refreshed, version bumped (`telegraph-vN+1`) |
-| Edit to `scripts/_lib/bench_telegraph*.py` | `telegraph` | report refreshed in same PR |
-| Edit to any rung source (frugality / telegraph / rtk / A/B) | `value` | `task value` re-renders `docs/value.md` in same PR |
-| Edit to `internal/bench/corpora/rtk/commands.yaml` | `rtk` | `scripts/bench_rtk_savings.py` refreshed in same PR |
-| Edit to `dist/router.json` or any rule frontmatter `triggers:` | `router-telemetry` | `task value:telemetry:replay` refreshes attribution map; required before any Phase 4 / Phase 5 cut |
-
-A PR that touches any of the cadence triggers without refreshing the
-corresponding report is rejected by reviewer convention (no CI gate yet
-— the trigger surface is too small to warrant one).
-
-## Cost envelope (`rtk` corpus)
-
-8 commands × 2 arms (raw vs. rtk-filtered) = 16 local shell invocations
-per run. Zero API spend — pure local measurement. Wall-time ≈ 5–10 s on
-the maintainer's repo (`scripts/bench_rtk_savings.py --quiet`).
-
-## Cost envelope (`telegraph` corpus)
-
-10 prompts × 3 arms (`condensed` · `terse-control` · `uncondensed`) = 30
-Anthropic calls per run. Observed envelope on `claude-sonnet-4-5` (v1,
-2026-05-16): **$0.0805 actual** · 0 errors · realised carve-out share
-30.67 %.
-
-## Commands
-
-```bash
-task bench -- --telegraph                                  # full run
-task bench -- --telegraph --telegraph-max-prompts 1          # 1-prompt smoke
-task bench -- --telegraph --telegraph-dry-run --no-write     # offline shape
-```
-
-Cost-touched runs require an `ANTHROPIC_API_KEY` at
-`~/.event4u/agent-config/anthropic.key` (mode 600).
-
-## Cross-references
-
-- [`benchmark-corpus-spec.md`](contracts/benchmark-corpus-spec.md) —
-  per-prompt schema.
-- [`benchmark-report-schema.md`](contracts/benchmark-report-schema.md) —
-  per-report JSON / Markdown contract.
-- [`condensation-default-kill-criterion.md`](contracts/condensation-default-kill-criterion.md)
-  — how a published `telegraph-v<N>` report is read against the kill table.
-- `agents/roadmaps/step-16-telegraph-substance.md` Phase 1 — where the
-  telegraph corpus was authored.

package/docs/capability-matrix.md

@@ -1,33 +0,0 @@
-# Capability matrix — what works on which host
-
-> **Generated** by `scripts/generate_capability_matrix.py` — do NOT
-> hand-edit. Derived from the `generate_tools()` projection logic in
-> `condense.py` (each cell traces to a `generate_*` dispatcher call).
-> Drift-checked in CI (`--check`).
-
-Cells: **✅ native** (host consumes the artifact directly — symlink /
-native dir) · **🔁 adapter** (projected through a host-specific
-transform — `.mdc`, workflow, or an aggregated single file) · **— none**
-(no generator emits this artifact for this host).
-
-| Artifact | claude-code | claude-plugin | augment | cursor | windsurf | cline | gemini | copilot | claude-desktop |
-|---|---|---|---|---|---|---|---|---|---|
-| `rules` | ✅ native | — none | ✅ native | 🔁 adapter | 🔁 adapter | ✅ native | 🔁 adapter | 🔁 adapter † | — none |
-| `skills` | ✅ native | ✅ native | ✅ native | — none | — none | — none | — none | — none | — none |
-| `commands` | ✅ native | — none | ✅ native | 🔁 adapter | 🔁 adapter | — none | — none | — none | — none |
-| `subagents` | ✅ native | — none | — none | 🔁 adapter | 🔁 adapter | 🔁 adapter | — none | — none | — none |
-| `personas` | ✅ native | — none | ✅ native | ✅ native | — none | — none | — none | — none | — none |
-| `user-types` | ✅ native | — none | ✅ native | ✅ native | — none | — none | — none | — none | — none |
-| `hooks` | — none | ✅ native | — none | — none | — none | — none | — none | — none | — none |
-
-## How to read this
-
-- Projection is **intentionally asymmetric** — a `— none` cell is a
-  design choice, not a bug. Skills project natively only where a host
-  has a native skill surface; everywhere else the rules + commands
-  carry the behaviour.
-- `🔁 adapter` cells are real coverage through a host-native shape
-  (Cursor `.mdc`, Windsurf workflows, the aggregated `GEMINI.md`).
-- `†` marks an **install-time** surface the installer writes (e.g.
-  `.github/copilot-instructions.md`), not the `generate_tools()` path —
-  real coverage, different code path.

package/docs/case-studies/_template.md

@@ -1,60 +0,0 @@
-# Case study — `<short-title>`
-
-> **Step-12 Phase 7 L128–131 template.** Drop a copy into `docs/case-studies/` with filename `<YYYY-MM-DD>-<user-type>-<slug>.md`. Anonymise as needed; metrics must be real.
-
-## Subject
-
-- **User-type:** (consultant | creator | founder | finance | ops | gtm | developer)
-- **Tool host:** (claude-code | cursor | windsurf | copilot | claude.ai web | other)
-- **`agent-config` install:** `npx agent-config init --user-type=<type>`
-- **Anonymised:** yes / no (if yes — describe what was abstracted)
-
-## Workflow before
-
-One paragraph. What was the specific job-to-be-done? How was it solved before `agent-config` (which tool, how many steps, where did it stall)? Include **time-to-output** as a baseline metric.
-
-## Workflow after
-
-One paragraph. Same job, after install. Which **3–5 skills** carried the load? Which **commands** did the user actually invoke? Where did the agent ask vs. act?
-
-## Top-10 skill invocations (proof of non-dev workflow)
-
-Paste output of `task bench --history --top 10` (or the equivalent host log). The closure gate (L130) requires this list to contain **zero** of: `test`, `deploy`, `ci`. If it does, the case study counts as proof of non-dev workflow rather than disguised dev work.
-
-```
- 1. <skill-name>          (N invocations)
- 2. ...
- ...
-10. ...
-```
-
-## Quantified outcomes
-
-| Metric | Before | After | Δ |
-|---|---:|---:|---:|
-| Time-to-output (minutes) | | | |
-| Number of revision passes | | | |
-| Self-reported confidence (1–5) | | | |
-| <domain-specific KPI> | | | |
-
-Example: *"Brief drafting: 90 min → 25 min, 73 % time reduction."*
-
-## What broke
-
-Be specific. The case study is only useful if it lists at least **two real friction points** — skill misfires, rule false-positives, missing user-type tags, confusing description language. These feed back into the corpus + skill descriptions.
-
-1. ...
-2. ...
-
-## Verbatim quote (optional, with consent)
-
-> "...short quote that captures the value the user got, in their voice..."
-
-— `<role>, <anonymised company shape>`
-
-## Provenance & consent
-
-- **Interview / live walkthrough date:** YYYY-MM-DD
-- **Consent for publication:** signed / verbal-recorded / declined-attribution
-- **Author:** maintainer initials
-- **Reviewed by user before merge:** yes / no

package/docs/case-studies/frontend-design-positioning.md

@@ -1,86 +0,0 @@
-# Competitive positioning — `frontend-design` pack vs. an external reference suite
-
-> Ours-vs-theirs verdict per the `competitive-positioning` skill,
-> closing `road-to-frontend-design-intelligence` Phase 8. Evidence:
-> full-tree deep-dive of an external reference suite
-> (2026-06-07) vs. this
-> repo post-adoption (ADR-061).
-
-## 1. Frame
-
-- **Theirs:** "Make any AI agent produce professional UI" — a
-  searchable design-knowledge corpus + BM25 engine + generative asset
-  add-ons, installed per-tool via an npm CLI. Segment: any developer
-  using an AI coding tool.
-- **Ours:** a governed skill/rule/command suite where frontend design is
-  one orchestrated capability among many — audit → design → apply →
-  review → polish with hard gates, now **grounded in the same corpus**.
-  Segment: teams standardizing AI-agent behavior across tools.
-
-## 2. Axes
-
-- **Design-knowledge grounding** — the differentiated capability that
-  triggered this adoption.
-- **Orchestration & gates** — who controls when knowledge is applied.
-- **Accessibility depth** — audit method, not just data.
-- **Stack coverage** — per-framework Do/Don't knowledge.
-- **Token authoring** — DTCG system + generators.
-- **Generative assets** — logo/CIP/banner generation.
-- **Governance & provenance** — licensing, refresh, safety surfaces.
-
-## 3. Verdict table
-
-| Axis | Ours (post-adoption) | Theirs | Verdict | Adopt? | Rationale |
-|---|---|---|---|---|---|
-| Design-knowledge grounding | Same corpus (11 CSVs + 16 stack CSVs + reasoning map) behind a **generalized manifest engine** with confidence + evidence-gap on every output | Origin of the corpus; frontend-hardcoded scripts | **Parity-plus** | done | ADR-061 adoption; our engine adds structured filters, named retrievers, schema-agnostic manifests |
-| Orchestration & gates | Directive set with audit gate, microcopy placeholder lock, a11y severity gate, polish ceiling | Skill prose + conventions; no enforced gates | **Ours** | — | Their corpus now feeds our gates — the combination argument inverts |
-| Accessibility depth | `accessibility-auditor` (WCAG 2.2 AA method) + corpus a11y data (contrast-adjusted palettes, chart grades, 44pt rules) | Checklist rows + a11y-graded data, no audit method | **Ours** | done | Their data enriched our gate; method stays ours |
-| Stack coverage | 16 stack corpora via `--stack` (Vue gap closed) + stack-dispatched executor skills | Same 16 CSVs | **Parity** | done | Verbatim adoption, one search mechanism |
-| Token authoring | `design-tokens` skill — DTCG 3-layer + Python generate/validate/embed (no Node dep) | `.cjs` generators (Node required) | **Ours (slight)** | done | Port removes the Node runtime requirement; brand→token sync deferred (watch note) |
-| Generative assets | **Not shipped** — gated by `domain-adoption-policy` (watch note); would adapt into `pack-ai-video` | Gemini logo/CIP/icon/banner/social suite | **Theirs** | gated | Heavy deps (`google-genai`, Chrome, API key) + no demand signal; revisit on trigger |
-| Governance & provenance | ADR-061, ATTRIBUTION with SHA pins, owner + cadence per corpus, runtime-safety review, Apache §4b markings | MIT/Apache files present; no refresh discipline documented | **Ours** | — | The axis their 86k-star repo does not compete on |
-
-## 4. Invariants (moats)
-
-- **Orchestration & gates** — inversion: "would a consumer drop our suite
-  for theirs?" Only if they wanted raw knowledge without enforcement;
-  the placeholder lock, audit gate, and a11y floor do not exist upstream.
-- **Governance & provenance** — refresh cadence + SHA pins + safety
-  reviews are structural in this package; a knowledge dump cannot match
-  them without becoming a governed suite itself.
-
-## 5. Adoption queue
-
-Empty — every `adopt? = done` row landed in
-`road-to-frontend-design-intelligence` Phases 1–6. The single open
-upstream capability (generative suite) is gated, not queued
-(`agents/settings/contexts/domain-watch/generative-brand-assets.md`).
-
-## 6. Unknowns
-
-- Their npm CLI's 18-tool projection breadth vs. our generator's
-  supported set: we verified parity for **our** targets
-  (`.claude` / `.cursor` / `.clinerules` / `.windsurf` / `.augment` via
-  `task generate-tools`); no claim about the 13 other tools they list —
-  we do not target them.
-- Real-session quality delta of grounded vs. ungrounded briefs — the
-  council's change-my-mind anchor (≥10 real sessions) applies to our own
-  Phase 9 expansion, and equally to claiming superiority here.
-
-## Net
-
-A consumer installing **our package alone** now gets the
-external reference suite's design knowledge **plus** orchestration, hard gates,
-a11y method, token authoring without Node, and provenance discipline.
-The combination is unnecessary; the gated generative suite is the one
-deliberate exception.
-
-## Upstream-contribute consideration (Step 8.3)
-
-Improvements worth proposing back (MIT obligations honored either way):
-the BM25 dedupe (`core.py` / `slide_search_core.py` byte-identical),
-the Python token toolchain (drops their Node dependency), and the
-manifest/filters layer. **Decision: defer the upstream PR** until the
-engine has soaked through one internal release cycle — proposing an
-interface we might still move is noise for the maintainer. Revisit at
-the first quarterly corpus refresh.