reselect-figures

name	reselect-figures
description	Use when the user wants to change which figures are embedded in xhs.md and wechat.md. Shows the user the figures with importance scores from analysis/06-figures.md, lets them multi-select per platform, re-runs both renderers with new selections.
disable-model-invocation	false
allowed-tools	Bash, Read, Write, Edit, Agent

paperstudio: reselect-figures workflow

Hard rules (non-negotiable, enforced by tests)

These apply to every Agent dispatch in this skill. Stage-local exceptions must be stated inline with a one-line "why".

1. Provenance — every output line 1: <!-- generated: <ts> by <agent> (paperstudio v<ver>) -->
2. Idempotence — OUTPUT_PATH exists + no --force ⇒ skip; exists + --force ⇒ cp $f $f.bak.NN first, then dispatch.
3. Log — after every dispatch (success or fail), call log_dispatch <subagent> <output-path> <ok|failed>. Never on a skip.
4. Paths — paper root is ${PAPERS_ROOT}, resolved once from ${CLAUDE_PAPERS_ROOT:-$HOME/claude-papers/papers}. Never hard-code ~/claude-papers/papers/ in writes.
5. Failure — Agent failure: log failed, do NOT delete partial output, do NOT auto-retry, surface the actionable rerun command, continue independent stages.
6. Chat language — reply in the user's invocation language. Artifact language is governed by the prompt's LANG= input.

Full text + rationale: see paperstudio/skills/_shared/dispatch-rules.md.

---

Invoke after /paperstudio:study has produced analysis/06-figures.md plus the notes set. Both notes/xhs.md and notes/wechat.md are re-rendered with the new figure selections; their prior versions are backed up.

Optional flags:

• --reinterpret: re-run figure-interpreter first to refresh analysis/06-figures.md importance scores. Useful if the original interpretation was off.
• --paper <slug>: target a specific paper folder.

---

Stage 1: Setup

1.1 Resolve target paper

Resolve target paper folder

Source the shared helper and resolve which paper folder this invocation targets:

source $CLAUDE_PLUGIN_ROOT/scripts/lib/resolve-paper.sh
resolve_paper "$@"
# After: $PAPER_DIR, $PAPER_SLUG, $PAPER_AUTODETECTED are set.
# If $PAPER_AUTODETECTED is "true", the helper already printed a warning to stderr.

If resolve_paper returns non-zero, abort with the helper's stderr message. Verify required files:

• $PAPER_DIR/analysis/06-figures.md
• $PAPER_DIR/images/ directory (non-empty)
• $PAPER_DIR/notes/source.md
• $PAPER_DIR/notes/titles.md
• $PAPER_DIR/notes/xhs.md
• $PAPER_DIR/notes/wechat.md

Abort with helpful message if any are missing.

Set:

• FIGURES_MD=$PAPER_DIR/analysis/06-figures.md
• IMAGES_DIR=$PAPER_DIR/images
• XHS_PATH=$PAPER_DIR/notes/xhs.md
• WECHAT_PATH=$PAPER_DIR/notes/wechat.md
• SOURCE_PATH=$PAPER_DIR/notes/source.md
• TITLES_PATH=$PAPER_DIR/notes/titles.md
• PLUGIN_ROOT=${CLAUDE_PLUGIN_ROOT}

Source the log-dispatch helper and extract plugin version:

source $CLAUDE_PLUGIN_ROOT/scripts/lib/log-dispatch.sh
PLUGIN_VERSION=$(grep -m1 '"version"' $CLAUDE_PLUGIN_ROOT/.claude-plugin/plugin.json | sed -E 's/.*"version"[^"]*"([^"]+)".*/\1/')

1.2 Optionally re-interpret figures

If --reinterpret is set:

First, back up the existing 06-figures.md:

NN=1
while [ -e "$FIGURES_MD.bak.$NN" ]; do
  NN=$((NN + 1))
done
cp "$FIGURES_MD" "$FIGURES_MD.bak.$NN"

Then dispatch figure-interpreter:

Agent(
  description: "figure-interpreter (refresh scores)",
  subagent_type: "general-purpose",
  prompt: <contents of $PLUGIN_ROOT/prompts/figure-interpreter.md> + concrete inputs:
    PAPER_TEXT=$PAPER_DIR/paper.txt (or paper.pdf if .txt missing)
    IMAGES_DIR=$IMAGES_DIR
    OUTPUT_PATH=$FIGURES_MD
    TEMPLATE_PATH=$PLUGIN_ROOT/templates/analysis/06-figures.md
    PLUGIN_VERSION=$PLUGIN_VERSION
)

Wait for completion.

log_dispatch figure-interpreter analysis/06-figures.md ok

If --reinterpret is not set, skip this step.

---

Stage 2: Show figures to user

2.1 Parse 06-figures.md

Read $FIGURES_MD. Parse YAML frontmatter to extract the figures: list — for each figure, capture file, caption, importance, role.

If the YAML is the FAILED placeholder (<!-- FAILED: ... -->), abort: "analysis/06-figures.md is FAILED — re-run /paperstudio:rerun-stage analysis or /paperstudio:reselect-figures --reinterpret first."

2.2 Display the menu

Show the user (in their invocation language):

Figures available (from analysis/06-figures.md):

  [1] page_1_img_1.png  importance=0.95  role=architecture
      Caption: "Overview of the model showing input → encoder → decoder."

  [2] page_3_img_2.png  importance=0.85  role=main-result
      Caption: "BLEU on WMT'14: our method 28.4 vs prior 27.3."

  [3] ...

Pick figures:
  - For xhs (1 figure): type a number, e.g. "1"
  - For wechat (2-3 figures): type comma-separated numbers, e.g. "1,2,4"
  - Type 'cancel' to abort.

xhs choice:

Wait for user input. Validate:

• xhs: exactly 1 number, must be in range [1, len(figures)].
• Re-prompt on invalid.

Then prompt for wechat:

wechat choice (2-3 numbers, comma-separated):

Wait for user input. Validate:

• wechat: 2 or 3 numbers, all in range, no duplicates.
• Re-prompt on invalid.

Set:

• XHS_FIGURES=[<images_dir>/<filename> for the chosen xhs figure] (1 path)
• WECHAT_FIGURES=[<images_dir>/<filename> for each chosen wechat figure] (2-3 paths)

If user types 'cancel' at either prompt, exit gracefully without modifying anything.

---

Stage 3: Backup + dispatch renderers

3.1 Back up both renderings

for path in "$XHS_PATH" "$WECHAT_PATH"; do
  NN=1
  while [ -e "$path.bak.$NN" ]; do
    NN=$((NN + 1))
  done
  cp "$path" "$path.bak.$NN"
done

3.2 Dispatch xhs-renderer + wechat-renderer in parallel

Issue both Agent calls in one message:

Agent(  // xhs
  description: "xhs-renderer with new figure selection",
  subagent_type: "general-purpose",
  prompt: <contents of $PLUGIN_ROOT/prompts/xhs-renderer.md> + concrete inputs:
    SOURCE_PATH=$SOURCE_PATH
    TITLES_PATH=$TITLES_PATH
    OUTPUT_PATH=$XHS_PATH
    TEMPLATE_PATH=$PLUGIN_ROOT/templates/notes/xhs.md
    SELECTED_FIGURES=<XHS_FIGURES>
    PLUGIN_VERSION=$PLUGIN_VERSION
)

Agent(  // wechat
  description: "wechat-renderer with new figure selection",
  subagent_type: "general-purpose",
  prompt: <contents of $PLUGIN_ROOT/prompts/wechat-renderer.md> + concrete inputs:
    SOURCE_PATH=$SOURCE_PATH
    TITLES_PATH=$TITLES_PATH
    OUTPUT_PATH=$WECHAT_PATH
    TEMPLATE_PATH=$PLUGIN_ROOT/templates/notes/wechat.md
    SELECTED_FIGURES=<WECHAT_FIGURES>
    PLUGIN_VERSION=$PLUGIN_VERSION
)

(Note: EDIT_INSTRUCTION and EXISTING_PATH are intentionally omitted — this skill re-renders from scratch with new figures. The user's prior body edits will not be preserved. If the user wants to preserve body edits, they should use /paperstudio:refine-notes with a figure-swap instruction instead.)

After both complete:

log_dispatch xhs-renderer notes/xhs.md ok
log_dispatch wechat-renderer notes/wechat.md ok

If either failed: log with failed status.

3.3 Verify outputs

If either output is missing or empty, restore from backup:

for path in "$XHS_PATH" "$WECHAT_PATH"; do
  if [ ! -s "$path" ]; then
    bak=$(ls -t "$path".bak.* 2>/dev/null | head -1)
    if [ -n "$bak" ]; then
      cp "$bak" "$path"
      echo "WARN: renderer produced empty output for $(basename "$path"); restored from $bak"
    fi
  fi
done

---

Stage 4: Final summary

✓ Reselected figures for both renderings.
  xhs.md:    <new figure list> (was: <prior figure list>)
  wechat.md: <new figure list> (was: <prior figure list>)
  Backups:
    notes/xhs.md.bak.<NN>
    notes/wechat.md.bak.<NN>
    <if --reinterpret was set: analysis/06-figures.md.bak.<NN>>

---

Notes

• Body content is regenerated. Unlike /paperstudio:refine-notes which surgically edits, /paperstudio:reselect-figures re-renders both files from source.md. Prior body edits (e.g., from earlier /paperstudio:refine-notes rounds) will be lost. To preserve body edits, use /paperstudio:refine-notes <platform> and ask the renderer to swap a figure as part of the instruction.
• --reinterpret only refreshes scores, doesn't change image files. The actual image files in images/ are produced by claude-paper:study and are read-only here.
• Translation: chat-facing prose to user is in user's invocation language. Renderings stay Chinese.