Discuss Claims

Discuss Claims

Post self-contained GitHub issues for internal team discussion of core empirical claims. Each issue covers one claim: motivation, methods, findings (with inline figure/table links), interpretation, implications, and limitations.

When to Use

• User wants to share analysis findings with co-authors via GitHub Issues
• User says "post issues", "create claim issues", "discuss on GitHub", "set up review"
• Analysis pipeline is complete and output files exist

When NOT to Use

• User wants a formal reviewer template with feedback/decision boxes (this skill creates discussion threads, not review forms)
• Repo is public and findings are embargoed
• Output files have not been generated yet

Workflow

Phase 0: Pre-flight

Run these checks before touching GitHub. Fix any failures before proceeding.

# 1. Issues enabled?
gh repo view OWNER/REPO --json hasIssuesEnabled

# 2. Existing [DISCUSS] issues?
gh issue list --state open | grep "\[DISCUSS\]"

# 3. Find untracked output files and auto-commit them
git ls-files --others --exclude-standard output/
# If any exist: git add output/ && git commit -m "Add output files for co-author review" && git push

CRITICAL — broken links are the #1 failure mode: GitHub blob links return 404 if the file is not committed and pushed. Always verify with git ls-files --error-unmatch for at least a sample of referenced figures/tables before creating any issue.

Phase 1: Collect Project Metadata

Infer from git or ask the user:

• GitHub repo: owner/repo (infer from git remote get-url origin)
• Branch: master or main (infer from git symbolic-ref refs/remotes/origin/HEAD)
• Output directories: where figures and tables live (default: output/figures/, output/tables/)

Construct the base blob URL:

https://github.com/OWNER/REPO/blob/BRANCH/PATH/TO/file.png

Phase 2: Discover Output Files

Scan the output directories and build a menu for the user to pick from per claim:

ls output/figures/*.png   # PNG only — PDFs do NOT render inline on GitHub
ls output/tables/*.csv
ls output/reports/*.md
ls code/                  # analysis scripts

PNG rule: Always link .png files, never .pdf. GitHub renders PNG inline; PDFs show as download links only.

Phase 3: Collect Claim Content (interactive, repeat N times)

Ask the user how many claims there are. Then for each claim, collect via AskUserQuestion or open-ended prompts:

1. 1-sentence claim statement (becomes the issue title: [DISCUSS] Claim N: ...)
2. Motivation: Why this analysis? What gap or puzzle does it address?
3. Method: What is the estimand? What design/scripts were used?
4. Key numbers: The 3–5 most important statistics
5. Figures + tables: User picks from the discovered file list (Phase 2)
6. Interpretation: What does the finding mean substantively?
7. Implications: Why does it matter for the paper's contribution?
8. Limitations (optional): Caveats and open questions

Phase 4: Create Master Tracker (Issue #1)

Always create this first so it gets the lowest issue number. Body contains:

• One-paragraph project description
• Navigation table: | Issue | Claim | One-line summary |
• Links to findings report, scripts dir, figures dir, tables dir

No labels. No milestone. No reviewer instructions.

Phase 5: Create Claim Issues (Issues #2 onward)

For each claim, create one issue using this structure:

## Motivation
[Why this analysis / what gap it fills]

## Data & Method
[Estimand, design, identification strategy]
**Scripts**: [script1.R](blob-url) · [script2.R](blob-url)

## Findings
[Key numbers in a table or bullets]

**Figures**:
- [figXX.png](blob-url) — [what to look at]

**Tables**: [tabXX.csv](blob-url) · ...
**Script reports**: [report.md](blob-url) · ...

## Interpretation
[What the pattern means]

## Implications
[Why it matters for the paper]

## Limitations & Open Questions
[Caveats, future directions — optional but encouraged]

Never include: labels, milestones, reviewer feedback sections, decision boxes, or tagging conventions like [SUBSTANTIVE].

Phase 6: Output

1. Print a summary table of all created issues with URLs.
2. Open the tracker in the browser: gh issue view 1 --repo OWNER/REPO --web

Quick Reference

| Step | Command | |------|---------| | Check Issues enabled | gh repo view OWNER/REPO --json hasIssuesEnabled | | Find untracked outputs | git ls-files --others --exclude-standard output/ | | Auto-commit outputs | git add output/ && git commit -m "Add output files" && git push | | Verify file tracked | git ls-files --error-unmatch output/figures/fig.png | | Create issue | gh issue create --repo OWNER/REPO --title "..." --body "..." | | Remove labels if present | gh issue edit N --repo OWNER/REPO --remove-label "label1,label2" | | Open in browser | gh issue view N --repo OWNER/REPO --web |

Common Mistakes

| Mistake | Fix | |---------|-----| | Link to untracked files → 404 | git ls-files --error-unmatch on sample files; auto-commit output/ first | | Use PDF links | Switch to .png — PDFs don't render inline on GitHub | | Include reviewer structure | This skill makes discussion threads: no feedback boxes, no decision fields | | Create claim issues before tracker | Tracker must be first to get the lowest issue number | | Add labels | Never add labels — always create clean, unlabeled issues | | Use raw GitHub URLs | Use blob URLs — renders PNG inline and CSV as formatted table |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: letitbk
• Source: letitbk/claude-academic-setup
• License: MIT

Install and usage instructions live in the source repository linked above.