@ericrisco/rsc 0.1.14 → 0.1.16

package/README.md

@@ -71,7 +71,7 @@ Run it inside any project and describe what you want. Working on the catalog
 itself? Clone and link:
 
 ```bash
-git clone https://github.com/ericrisco/skills.git ~/rsc-skills
+git clone https://github.com/ericrisco/rsc-harness.git ~/rsc-skills
 cd ~/rsc-skills && npm install && npm link
 ```
 

package/manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "0.1.14",
+  "version": "0.1.16",
   "counts": {
     "skills": 231
   },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ericrisco/rsc",
-  "version": "0.1.14",
+  "version": "0.1.16",
   "description": "Eric Risco's agent-skills catalog as a granular, self-recommending CLI installer.",
   "type": "module",
   "bin": {
@@ -47,13 +47,13 @@
   ],
   "license": "MIT",
   "author": "Eric Risco",
-  "homepage": "https://github.com/ericrisco/skills#readme",
+  "homepage": "https://github.com/ericrisco/rsc-harness#readme",
   "repository": {
     "type": "git",
-    "url": "git+https://github.com/ericrisco/skills.git"
+    "url": "git+https://github.com/ericrisco/rsc-harness.git"
   },
   "bugs": {
-    "url": "https://github.com/ericrisco/skills/issues"
+    "url": "https://github.com/ericrisco/rsc-harness/issues"
   },
   "publishConfig": {
     "access": "public"

package/skills/harness/SKILL.md

@@ -11,7 +11,7 @@ profiles: [minimal, core, full]
 The **harness** is the control plane of a workspace. A workspace need not be code: it can be a company, an ops desk, a legal archive, a personal knowledge vault. Whatever it is, the harness is the durable apparatus that keeps it operable and legible, made of three parts:
 
 - **`01-TOOLS/<PROVIDER>/`** — the operational tooling layer. One folder per external provider, co-locating credentials (`.env`) with the scripts that consume them. Each tool ships a working `test_connection` against the real API.
-- **`02-DOCS/`** — the **Karpathy chaos→knowledge engine**: a domain-agnostic LLM wiki (`inbox/` + `raw/` + `raw/worklog/` + `wiki/` + `wiki/index.md` + `wiki/log.md` + `wiki/gaps.md` + `wiki/scores.json` + the `.base` views), fully embedded in this skill. It feeds from **two on-ramps**: (1) the user drops **any raw file in any format** (PDF, image, CSV, JSON, txt, html…) into `inbox/`, and an **Inbox Sweep** ("the agent goes for a walk") extracts, classifies, cross-links, and compiles it; (2) a **Worklog Sweep** captures **what we do** — every meaningful session of work is itself a `raw` source (`raw/worklog/`) compiled into the wiki, fired by a `PreCompact`/`SessionEnd` hook, by an explicit milestone (a commit), or by the daily curation automation (`references/daily-curation-automation.md`). The wiki is **Obsidian-native**: wikilinks, YAML frontmatter (Properties), readable filenames, and `.base` views give a real graph + live tables — structure, **not vector DB / embeddings / RAG**. The agent writes it; the human reads it in Obsidian. See `references/wiki-protocol.md` for the protocol, `references/ingest-formats.md` for the multiformat Fetch, `references/wiki-worklog-template.md` for the work capture, and `references/obsidian-scaffolding.md` for the vault scaffolding. Topics are inferred from content (`finanzas/`, `legal/`, `crm/`…), never hardcoded. The wiki **self-improves continuously**: every Ingest, Sweep and Query triggers a Maintenance Pass (deterministic lint, score recomputation, gap detection, Related/See-Also sweep), and every N interactions a Micro-Improve runs. Deep Improve runs on explicit request or via the scheduled daily curation. No external skill needed.
+- **`02-DOCS/`** — the **Karpathy chaos→knowledge engine**: a domain-agnostic LLM wiki (`inbox/` + `raw/` + `raw/worklog/` + `wiki/` + `wiki/index.md` + `wiki/log.md` + `wiki/gaps.md` + `wiki/scores.json` + the `.base` views), fully embedded in this skill. It feeds from **two on-ramps**: (1) the user drops **any raw file in any format** (PDF, image, CSV, JSON, txt, html…) into `inbox/`, and an **Auto-Ingest Sweep** ("the agent goes for a walk") extracts, classifies, cross-links, and compiles it — and goes further, **discovering un-ingested documents anywhere in the workspace** (a stray `/transcripts`, `/facturas`…), bounded by `.rscignore` and de-duplicated by the `wiki/.ingested.json` ledger, fired automatically at SessionStart (when the inbox has material) and by the daily cron. Ingest is non-destructive (originals are copied, never moved); removing an emptied folder needs explicit consent; (2) a **Worklog Sweep** captures **what we do** — every meaningful session of work is itself a `raw` source (`raw/worklog/`) compiled into the wiki, fired by a `PreCompact`/`SessionEnd` hook, by an explicit milestone (a commit), or by the daily curation automation (`references/daily-curation-automation.md`). The wiki is **Obsidian-native**: wikilinks, YAML frontmatter (Properties), readable filenames, and `.base` views give a real graph + live tables — structure, **not vector DB / embeddings / RAG**. The agent writes it; the human reads it in Obsidian. See `references/wiki-protocol.md` for the protocol, `references/ingest-formats.md` for the multiformat Fetch, `references/wiki-worklog-template.md` for the work capture, and `references/obsidian-scaffolding.md` for the vault scaffolding. Topics are inferred from content (`finanzas/`, `legal/`, `crm/`…), never hardcoded. The wiki **self-improves continuously**: every Ingest, Sweep and Query triggers a Maintenance Pass (deterministic lint, score recomputation, gap detection, Related/See-Also sweep), and every N interactions a Micro-Improve runs. Deep Improve runs on explicit request or via the scheduled daily curation. No external skill needed.
 - **The Knowledge map** — the `## Knowledge map` section of the root `CLAUDE.md` that indexes the wiki (including the `harness/` topic) and is read by every other skill before it works in its area.
 
 `harness` is the **protagonist concept**. `init` is the bootstrap front door — it gauges the user, drafts the profile, and hands off the first scaffold. THIS skill (`harness`) is the **ongoing control**: it audits, migrates, scaffolds, sweeps the inbox, and keeps the wiki, the tooling and the Knowledge map honest over the life of the workspace. It also generates root `CLAUDE.md` and `AGENTS.md`, and migrates legacy `XX-*` numbered folders into the canonical layout.

package/skills/harness/references/daily-curation-automation.md

@@ -14,7 +14,10 @@ cron via the `schedule` skill pointed at this prompt.
 
 Run once per day, at a quiet hour, to keep the second brain alive:
 
-- process pending `raw/worklog/` and `inbox/` material into `wiki/`
+- run the **Auto-Ingest Sweep**: process `inbox/`, then scan the workspace (minus
+  `.rscignore`) for un-ingested documents, ingest the clearly-documentary folders,
+  record them in `wiki/.ingested.json`, and propose the ambiguous ones
+- process pending `raw/worklog/` material into `wiki/`
 - strengthen weak / low-score pages; fill one open gap
 - repair links and the graph (resolve broken wikilinks, add `## Related` where proper nouns recur)
 - mark `stale` pages whose sources moved on
@@ -26,17 +29,21 @@ Daily, at a quiet hour when the user is unlikely to be editing the vault.
 
 ## Scope (paths)
 
-`raw/worklog/`, `inbox/`, `raw/`, `wiki/`, `wiki/harness/`.
+The **workspace root** (for Auto-Ingest discovery, bounded by `.rscignore`),
+`raw/worklog/`, `inbox/`, `raw/`, `wiki/`, `wiki/.ingested.json`, `wiki/harness/`.
 
 ## Prompt
 
 ```text
 Run the daily curation pass for this 02-DOCS second brain.
 
-Use the harness wiki-protocol.md as the contract. Compile pending raw/worklog/
-and inbox/ material into wiki/ pages (update existing before creating new),
-keep frontmatter + wikilinks consistent, append significant decisions to
-wiki/harness/decisions.md, refresh the .base-backed navigation, recompute
+Use the harness wiki-protocol.md as the contract. First run the Auto-Ingest Sweep:
+process inbox/, then scan the workspace (minus .rscignore) for un-ingested documents,
+ingest the clearly-documentary folders (copy originals, never move), record them in
+wiki/.ingested.json, and list ambiguous folders as proposals (do not grab them).
+Then compile pending raw/worklog/ material into wiki/ pages (update existing before
+creating new), keep frontmatter + wikilinks consistent, append significant decisions
+to wiki/harness/decisions.md, refresh the .base-backed navigation, recompute
 scores.json + the score: property, and append a dated entry to wiki/log.md.
 
 Be conservative. If the topic map no longer fits, record a recommendation in

package/skills/harness/references/ingest-ignore-defaults.md

@@ -0,0 +1,48 @@
+# Ignore defaults — what the Auto-Ingest Sweep must never scan
+
+The Auto-Ingest Sweep discovers un-ingested documents across the whole workspace.
+`02-DOCS/.rscignore` scopes that scan so it never grabs code, build output, tooling,
+the wiki's own internals, or non-document files. This is the single guardrail that
+keeps "automatic, workspace-wide" from turning into "tragar lo que no toca".
+
+`02-DOCS/.rscignore` is **tracked** (not gitignored): the team shares the same scan
+boundary. Users append their own patterns freely.
+
+## Default `02-DOCS/.rscignore`
+
+```gitignore
+# 02-DOCS/.rscignore — paths the Auto-Ingest Sweep must NEVER scan as a knowledge source.
+# Tracked (the team shares it). Append your own patterns below.
+node_modules/
+.git/
+.next/
+dist/
+build/
+.venv/
+__pycache__/
+.dart_tool/
+.pytest_cache/
+.ruff_cache/
+01-TOOLS/
+02-DOCS/inbox/_processed/
+02-DOCS/raw/
+02-DOCS/wiki/
+.DS_Store
+# Only document-like files are candidates anyway — pdf, docx, doc, md, txt, csv,
+# tsv, json, html, rtf, and doc-like images (png/jpg of scans). Everything else
+# (code, archives, binaries, lockfiles) is ignored even outside these paths.
+```
+
+## App source dirs are ignored at scan time
+
+Beyond the file above, the sweep adds every detected **app source directory** to the
+effective ignore set: any directory carrying a manifest (`package.json`,
+`pyproject.toml`, `pubspec.yaml`, `go.mod`, `Cargo.toml`) and its tree. Source code is
+never a knowledge source — `02-DOCS/wiki/` is where understanding about the code lives,
+written deliberately, not scraped from the repo.
+
+## Why `02-DOCS/raw/` and `02-DOCS/wiki/` are ignored
+
+They are *outputs* of ingestion, not inputs. Scanning them would re-ingest the wiki
+into itself. `inbox/` is the only `02-DOCS/` subtree the sweep reads as a source
+(plus discovered external folders).

package/skills/harness/references/wiki-protocol.md

@@ -17,6 +17,7 @@ skill required. When the parent skill (`SKILL.md`) reaches Phase 4 step 8
 - [Ingest](#ingest) · [Fetch](#fetch-raw) · [Compile](#compile-wiki) · [Cascade Updates](#cascade-updates) · [Post-Ingest](#post-ingest)
 - [Inbox Sweep — "el agente sale a pasear"](#inbox-sweep--el-agente-sale-a-pasear)
 - [Worklog Sweep — work-driven capture](#worklog-sweep--work-driven-capture)
+- [Auto-Ingest Sweep — automatic, workspace-wide](#auto-ingest-sweep--automatic-workspace-wide)
 - [Query](#query)
 - [Lint](#lint) · [Deterministic Checks](#deterministic-checks-auto-fix) · [Heuristic Checks](#heuristic-checks-report-only)
 - [Continuous Improvement](#continuous-improvement) — Maintenance Pass / Micro-Improve / Deep Improve
@@ -172,6 +173,9 @@ exist. Create only what is missing; never overwrite existing files:
 - `02-DOCS/wiki/log.md` — heading `# Wiki Log`, empty body
 - `02-DOCS/wiki/gaps.md` — heading `# Knowledge Gaps`, empty body
 - `02-DOCS/wiki/scores.json` — `{}` (populated by the first Maintenance Pass)
+- `02-DOCS/wiki/.ingested.json` — `{}` (the Auto-Ingest Sweep's seen-ledger)
+- `02-DOCS/.rscignore` — the scan boundary for the Auto-Ingest Sweep, from
+  `ingest-ignore-defaults.md` (tracked, not gitignored)
 - `02-DOCS/wiki/reports/` directory (with `.gitkeep`) — will hold Deep Improve reports
 - `02-DOCS/audits/` directory (with `.gitkeep`) — holds per-run audit reports
   from the parent skill (`audit-YYYY-MM-DD-HHMM.html`). May already exist if
@@ -426,6 +430,62 @@ Append to `02-DOCS/wiki/log.md`:
 
 ---
 
+## Auto-Ingest Sweep — automatic, workspace-wide
+
+The Inbox Sweep processes what the user *dropped* in `inbox/`. The **Auto-Ingest
+Sweep** extends it: it also **discovers** un-ingested documents anywhere in the
+workspace and ingests them on its own — so the user never has to remember to file
+sources. It runs automatically (SessionStart nudge + daily cron; see Triggers) and
+on request.
+
+The hard rule that keeps "automatic, workspace-wide" safe: **ingesting is
+non-destructive** (read + create wiki pages + copy the original into
+`raw/<topic>/_originals/`), so it is safe to do unattended. **Moving or deleting
+anything is destructive** and always requires explicit, quoted consent.
+
+### Steps
+
+1. **inbox/** — run the Inbox Sweep on `inbox/` (unchanged).
+2. **Scan** — walk the workspace minus `.rscignore` (and minus detected app source
+   dirs). Consider only document-like files. A file is **un-ingested** when its
+   `path` is absent from `wiki/.ingested.json` OR its content hash differs.
+3. **Classify** each candidate folder:
+   - **Clearly documentary** (a folder of docs/transcripts/pdfs, no code) →
+     **auto-ingest**: Compile to `wiki/`, copy the original into
+     `raw/<topic>/_originals/` (**copy, never move** — the source stays put).
+   - **Ambiguous** (mixed, or possibly app data) → **list it and propose**; do not
+     touch it. Log it to `gaps.md` so the next pass remembers.
+4. **Record** every ingested source in `wiki/.ingested.json`.
+5. **Offer cleanup** — for a folder the sweep fully ingested, **offer** to remove it
+   once empty, with the consent quoted back from the user. **Never auto-delete**, and
+   never a folder the sweep did not itself empty.
+
+### Ledger — `wiki/.ingested.json`
+
+The seen-ledger; the single source of truth for "what we've already ingested",
+which makes the workspace scan idempotent (the Inbox Sweep alone tracks
+processed-ness only by moving files to `inbox/_processed/`).
+
+```json
+{
+  "<relpath-from-workspace-root>": { "hash": "sha256:…", "ingested": "YYYY-MM-DD", "topic": "…" }
+}
+```
+
+Un-ingested = path absent OR hash changed. `inbox/` files still move to
+`inbox/_processed/` **and** get a ledger entry. Initialized as `{}`.
+
+### Auto-Ingest safety rails
+
+- **Ledger-idempotent.** Never reprocess a source whose path+hash is recorded.
+- **Scoped by `.rscignore`.** Never scan code, build output, `01-TOOLS/`, the
+  wiki's own `raw/`+`wiki/`, or non-document files (see `ingest-ignore-defaults.md`).
+- **Copy, don't move.** Discovered originals are copied; the source is left in place.
+- **Consent for destruction.** Removing an emptied folder is quoted-consent only.
+- **Ambiguity is proposed, not grabbed.**
+
+---
+
 ## Query
 
 Search the wiki and answer questions. Examples of triggers:

package/targets/session-start.sh

@@ -1,8 +1,9 @@
 #!/usr/bin/env bash
 # rsc SessionStart payload (claude). Wired by targets/claude.js.
 #   $1 = absolute path to suggest's SKILL.md   $2 = absolute project root
-# Always emits suggest's always-on body; appends an onboarding banner only when
-# the workspace has no harness profile yet and the user has not opted out.
+# Always emits suggest's always-on body; appends an onboarding banner when the
+# workspace has no harness profile yet, and an auto-ingest nudge when there is
+# un-ingested material waiting in the inbox.
 set -u
 
 cat "$1" 2>/dev/null
@@ -20,3 +21,23 @@ If the user does not want a harness here: create .rsc/.no-harness
 ==========================
 BANNER
 fi
+
+# Auto-Ingest nudge: when a harness wiki exists and the inbox holds a real file
+# (anything other than README.md / dotfiles / the _processed archive), tell the
+# agent to run the Auto-Ingest Sweep. The hook only reminds; the agent does the
+# scan + ingest. Cheap signal here; the thorough workspace scan lives in the sweep.
+inbox="$2/02-DOCS/inbox"
+if [ -d "$2/02-DOCS/wiki" ] && [ -d "$inbox" ]; then
+  pending=$(find "$inbox" -maxdepth 1 -type f ! -name 'README.md' ! -name '.*' 2>/dev/null | head -1)
+  if [ -n "$pending" ]; then
+    cat <<'BANNER'
+
+===== rsc auto-ingest =====
+Un-ingested material is waiting in 02-DOCS/inbox/.
+ACTION: run the Auto-Ingest Sweep now — ingest inbox/, then scan the workspace
+(minus .rscignore) for un-ingested documents, recording them in wiki/.ingested.json.
+Originals are copied, never moved; deleting an emptied folder needs explicit consent.
+===========================
+BANNER
+  fi
+fi