docs(knowledge): fix youtube-digest contact-sheet retention wording + --target resolution gap

plugins/knowledge/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/claude-code-plugin-manifest.json",
   "name": "knowledge",
-  "version": "0.9.2",
+  "version": "0.9.3",
   "description": "Ingest external knowledge into durable, synthesized artifacts. Ships a book-distillation pipeline (PDF/EPUB into concept-organized, author-attributed skill reference files), a YouTube pipeline (watch, transcript, link harvest, and repo-applicability synthesis), and a course-digest pipeline (extract and synthesize online video courses — Dometrain, Teachable — into repo-applicable recommendations), plus a re-runnable setup action; a configurable library directory governs where synthesized artifacts land in the consuming repo.",
   "author": {
     "name": "Melodic Software",

plugins/knowledge/CHANGELOG.md

@@ -4,6 +4,23 @@ All notable changes to the `knowledge` plugin are recorded here. The `version` i
 `.claude-plugin/plugin.json` is the delivery vehicle — a consumer receives a change
 only after that version increases.
 
+## [0.9.3]
+
+### Fixed
+
+- **youtube-digest: contact-sheet retention wording + `--target` resolution gap**
+  (#1015): the intro paragraph called the `key-frames/contact-sheets/` snapshot
+  "temp-only handling," contradicting the Output contract's "local DR snapshot,
+  gitignored" characterization of the same directory; reworded to "never-committed
+  handling" so the snapshot reads as a durable-on-disk-but-gitignored instance, not
+  temp state. `--target <repo>` resolution now requires a **local working tree on
+  disk** (not just a name) because `templates/synthesis-item.md`'s grep-backed
+  **Target touchpoints** need a tree to grep; an explicit `--target` with no local
+  checkout now halts and asks for its path instead of falling through to
+  `CLAUDE_PROJECT_DIR`/CWD or inventing paths, and both the target name and its
+  resolved local tree path are recorded in `README.md`'s `**Target:**` line so
+  `resume` re-greps the same tree.
+
 ## [0.9.2]
 
 ### Fixed

plugins/knowledge/skills/youtube-digest/SKILL.md

@@ -16,7 +16,7 @@ ImageMagick: !`magick -version 2>/dev/null | head -1 || echo "MISSING — instal
 
 # YouTube
 
-Absorb a single public YouTube video (transcript + visual frames), harvest reference links, fact-check claims through deeper external research, and produce a prioritized repo-applicability menu. Video download, bulk frames, working contact sheets, and shallow git clones use the OS temp directory — durable artifacts live under `.work/<watch-epic>/<video-slug>/` per the video-digest slice convention. One contact-sheet exception: `snapshot-bootstrap.js` copies the sheets into the slice at `key-frames/contact-sheets/*.jpg` as a **local disaster-recovery snapshot** — durable on disk (it survives `tempSession` cleanup, so recovering the sheets does not require re-running acquisition) but gitignored, so it is never committed (see the Output contract). That temp-only handling is otherwise a fixed, non-configurable part of this contract: a consumer that wants the source video, bulk frames, or contact sheets retained as a **committed**, re-runnable substrate (rather than regenerable temp state or a gitignored local snapshot) does not get that from this skill today — a documented, LFS-aware retention path is a tracked follow-up, not yet built.
+Absorb a single public YouTube video (transcript + visual frames), harvest reference links, fact-check claims through deeper external research, and produce a prioritized repo-applicability menu. Video download, bulk frames, working contact sheets, and shallow git clones use the OS temp directory — durable artifacts live under `.work/<watch-epic>/<video-slug>/` per the video-digest slice convention. One contact-sheet exception: `snapshot-bootstrap.js` copies the sheets into the slice at `key-frames/contact-sheets/*.jpg` as a **local disaster-recovery snapshot** — durable on disk (it survives `tempSession` cleanup, so recovering the sheets does not require re-running acquisition) but gitignored, so it is never committed (see the Output contract). That never-committed handling is a fixed, non-configurable part of this contract: a consumer that wants the source video, bulk frames, or contact sheets retained as a **committed**, re-runnable substrate (rather than regenerable temp state or a gitignored local snapshot) does not get that from this skill today — a documented, LFS-aware retention path is a tracked follow-up, not yet built.
 
 Where this skill says "deeper research," use whatever external-research capability your project provides — for example the discovery plugin's `/discovery:research` / `/discovery:research-deep` when installed. Treat those as the reference implementation, not a hard dependency.
 
@@ -286,7 +286,7 @@ node "${CLAUDE_PLUGIN_ROOT}/skills/youtube-digest/extraction/run.mjs" evals/chec
 
 1. **Synthesis** (after research gate) — template: `templates/synthesis-item.md`
 
-   **Synthesis target resolution** — every menu item and `templates/readme-journey.md`'s TLDR are framed against one resolved target, never an implicit "the repo I'm in": explicit `--target <repo>` argument → the consuming project (`CLAUDE_PROJECT_DIR`) when `watch` is invoked directly inside a repo, with no separate corpus session → ask. Whichever rung resolves it, record the target name in `README.md`'s `**Target:**` line (`templates/readme-journey.md`) so a later `resume` or downstream consumer never has to re-infer it. This aligns with (but does not depend on) the `/knowledge:apply` design (`docs/knowledge-integration-design.md`), which will eventually take over repo-fitting via its own `--target` argument from a corpus session; when that skill is installed and built, prefer it for cross-repo fitting and treat this skill's menu as its input, not a replacement.
+   **Synthesis target resolution** — every menu item and `templates/readme-journey.md`'s TLDR are framed against one resolved target, never an implicit "the repo I'm in". Because `templates/synthesis-item.md`'s **Target touchpoints** are grep-backed, the target must resolve to a **local working tree on disk**, not merely a name: explicit `--target <repo>`, resolved to a local checkout of that repo → the consuming project (`CLAUDE_PROJECT_DIR`) when `watch` runs directly inside a repo, with no separate corpus session → ask. An explicit `--target <repo>` that has no local checkout (e.g. run from a separate corpus session where that repo isn't cloned locally) does **not** resolve — stop and ask for its local checkout path rather than falling through to `CLAUDE_PROJECT_DIR`, grepping the current directory, or inventing touchpoint paths. Whichever rung resolves it, record both the target name and its resolved local tree path in `README.md`'s `**Target:**` line (`templates/readme-journey.md`) so a later `resume` or downstream consumer re-greps the same tree instead of re-inferring it. This aligns with (but does not depend on) the `/knowledge:apply` design (`docs/knowledge-integration-design.md`), which will eventually take over repo-fitting via its own `--target` argument from a corpus session (auto-cloning the resolved repo); when that skill is installed and built, prefer it for cross-repo fitting and treat this skill's menu as its input, not a replacement.
 
    - Materialize `recommendations/` from `templates/recommendations/` (hub README links all docs)
    - `recommendations/menu.md` — categories: `immediate-takeaway` | `worth-investigating` | `poc-candidate` | `full-slice` | `no-go`; P0–P2 + consensus notes

plugins/knowledge/skills/youtube-digest/templates/readme-journey.md

@@ -6,7 +6,7 @@ updated: YYYY-MM-DDTHH:MM:SSZ
 
 # {Video Title}
 
-**Target:** {resolved synthesis target — SKILL.md "Synthesis target resolution"}
+**Target:** {resolved synthesis target — name + local tree path; SKILL.md "Synthesis target resolution"}
 
 ## TLDR