Octofund

Behavior

On First Interaction

1. Initialize SQL tracking tables (if not already created):

CREATE TABLE IF NOT EXISTS octofund_runs (
  id INTEGER PRIMARY KEY AUTOINCREMENT,
  budget REAL NOT NULL,
  projects_scored INTEGER NOT NULL,
  projects_funded INTEGER NOT NULL,
  run_date TEXT DEFAULT (date('now')),
  created_at TEXT DEFAULT (datetime('now'))
);
CREATE TABLE IF NOT EXISTS octofund_scores (
  id INTEGER PRIMARY KEY AUTOINCREMENT,
  run_id INTEGER NOT NULL,
  project TEXT NOT NULL,
  impact_score REAL,
  gap_score REAL,
  combined_score REAL,
  maintainers INTEGER,
  dependents TEXT,
  funding_monthly REAL,
  has_sponsors INTEGER DEFAULT 0,
  has_opencollective INTEGER DEFAULT 0,
  has_funding_yml INTEGER DEFAULT 0,
  has_tidelift INTEGER DEFAULT 0,
  has_security_md INTEGER DEFAULT 0,
  hidden_pillar INTEGER DEFAULT 0,
  suggested_amount REAL,
  override_amount REAL,
  excluded INTEGER DEFAULT 0,
  FOREIGN KEY (run_id) REFERENCES octofund_runs(id)
);

2. Ask for the budget amount if not provided:

Use `ask_user` tool:
"What's the budget to allocate?"
allow_freeform: true

3. Check for overrides — if the user says something like "pin curl at $3,000", parse and apply.

Data Collection

For each project in the seed list (data/critical-projects.json), read the file with view, then collect:

1. Dependents Count

gh api repos/{owner}/{repo} --jq '.network_count // 0, .forks_count // 0, .stargazers_count // 0, .watchers_count // 0'

Also check the dependency graph:

gh api repos/{owner}/{repo}/dependents 2>/dev/null || echo "unavailable"

Note: GitHub does not expose a direct "dependents count" API endpoint. When the API doesn't return this data, use the dependents count from the seed list's dependents_estimate field. If neither is available, show ⚠️ unavailable.

2. Funding Signals

GitHub Sponsors:

gh api graphql -f query='{ user(login: "{owner}") { hasSponsorsListing sponsorsListing { activeGoal { percentComplete } } } }' 2>/dev/null

If the owner is an org:

gh api graphql -f query='{ organization(login: "{owner}") { hasSponsorsListing } }' 2>/dev/null

FUNDING.yml:

gh api repos/{owner}/{repo}/contents/.github/FUNDING.yml --jq '.content' 2>/dev/null | base64 -d 2>/dev/null

OpenCollective:

curl -sf "https://opencollective.com/{project}/members/all.json" 2>/dev/null | head -c 200

If this returns data, the project has an OpenCollective. If 404 or empty, it doesn't.

Tidelift: Check if the project's package is listed. Use the ecosystem from the seed list:

curl -sf "https://tidelift.com/lifter/search/{ecosystem}/{package}" -o /dev/null -w "%{http_code}" 2>/dev/null

Security posture:

gh api repos/{owner}/{repo}/contents/SECURITY.md --jq '.name' 2>/dev/null
gh api repos/{owner}/{repo}/vulnerability-alerts --silent 2>/dev/null; echo $?

3. Maintainers Count

gh api repos/{owner}/{repo}/contributors --jq 'length' --paginate 2>/dev/null | head -1

Use the top contributors with recent commits (last 12 months) as "active maintainers." If unavailable, use the maintainers_estimate from the seed list.

Scoring

Read scoring.md with the view tool for the full rubric. Apply:

1. Impact Score (0–100): Normalize dependents count against the highest in the set. The project with the most dependents gets 100; others scale proportionally.

2. Funding Gap Score (0–100): Check 6 signals. Each absent signal adds points:

   • No GitHub Sponsors: +20
   • No OpenCollective: +15
   • No FUNDING.yml: +15
   • No Tidelift: +10
   • No SECURITY.md / advisories: +10 (informational — still counts toward gap)
   • Monthly funding $0: +30 | <$500: +20 | <$1K: +10 | >$1K: +0

3. Combined Score: (Impact × 0.6) + (Gap × 0.4)

4. Hidden Pillar Boost 🔥: If active maintainers ≤ 3, add +10 to combined score (cap at 100).

5. Rank: Sort by combined score descending.

6. Allocation:

   • Exclude projects scoring below 70.
   • Apply overrides first (pinned amounts).
   • Distribute remaining budget proportionally to combined score.
   • Enforce floor ($500) and ceiling (30% of total budget) per project.
   • Show unallocated remainder explicitly if rounding creates one.

Report Output — LOCKED TEMPLATE

CRITICAL: Produce this exact structure every run. Do not add, remove, or reorder sections.

# 🐙 OctoFund — ${BUDGET}
**{FUNDED} of {TOTAL} projects funded · {DATE}**

---

### Top 3

| | Project | Maintainers | Dependents | Funding | Suggested |
|---|---------|:-----------:|:----------:|:-------:|----------:|
| 🥇 | **{project}** 🔥? | {n} | {n}K | ${n}/mo | ${n} |
| 🥈 | **{project}** 🔥? | {n} | {n}K | ${n}/mo | ${n} |
| 🥉 | **{project}** 🔥? | {n} | {n}K | ${n}/mo | ${n} |

🔥 **Hidden Pillar** — ≤3 maintainers supporting critical infrastructure

---

### All Scores

| # | Project | Impact | Gap | Score | Maintainers |
|---|---------|:------:|:---:|:-----:|:-----------:|
(all projects, sorted by score descending)

**Impact** = how many repos depend on it
**Gap** = how underfunded relative to usage
**Score** = Impact (60%) + Gap (40%), with Hidden Pillar boost for ≤3 maintainers

---

### Funding Across Platforms

| Project | Sponsors | OpenCollective | FUNDING.yml | Tidelift | Est. $/mo |
|---------|:--------:|:--------------:|:-----------:|:--------:|----------:|
(all projects, same order as scores)

❌ None · 🟡 Present but $0 · 🟠 <$1K/mo · 🟢 >$1K/mo

---

### Recommended Allocation

| # | Project | Why fund | Amount |
|---|---------|----------|-------:|
(funded projects, then excluded with strikethrough)

| Budget | ${BUDGET} |
|--------|--------:|
| Allocated | ${ALLOCATED} |
| Projects funded | {n} of {n} |
| Floor / Ceiling | $500 / ${CEILING} |

---

> 📄 Export to CSV?

After Report

1. Store results in SQL:

INSERT INTO octofund_runs (budget, projects_scored, projects_funded) VALUES ({budget}, {scored}, {funded});
-- then for each project:
INSERT INTO octofund_scores (run_id, project, impact_score, gap_score, combined_score, ...) VALUES (...);

2. CSV Export: If user says yes, write octofund-{DATE}.csv to current directory with all columns.

3. Quarter comparison: If user asks to compare, query octofund_scores from previous runs.

Guardrails — DO NOT VIOLATE

Behavioral

1. Single purpose — This skill produces a funding report. It does NOT send emails, create issues, open PRs, or take any action beyond reporting. Future commit mode requires explicit user confirmation for every action.
2. No hallucinated data — Every number must come from an API call or the seed list. If an API call fails, show ⚠️ unavailable, never estimate or make up numbers.
3. No editorial opinions — The "Why fund" column states facts only (e.g., "1 maintainer, 34K dependents, $0 funding"). Never use subjective language.
4. Fixed template — The report sections (Top 3 → All Scores → Funding Across Platforms → Recommended Allocation) are locked. Do not add, remove, or reorder sections.

Input

5. Dollar amount required — If the user doesn't provide a budget, use ask_user to request one. Do not assume.
6. Minimum budget — Reject amounts below $500: "Minimum budget is $500 to ensure meaningful allocations."
7. Seed list only — Score projects from critical-projects.json plus any the user adds at runtime. Do not discover or suggest new projects autonomously.
8. Override validation — Pinned allocations must not exceed 30% of budget per project.

Output

9. Floor/ceiling enforced — No project gets less than $500 or more than 30% of total budget.
10. Unallocated remainder — If rounding leaves leftover, show it explicitly. Do not force-allocate.
11. Consistent emoji legend — Always use ❌🟡🟠🟢 as defined. No variations.
12. No run-to-run memory — Each invocation is independent. Do not reference or compare to previous runs unless the user explicitly asks.

Structural

13. Pure markdown skill — All logic lives in this SKILL.md. No external scripts, no shell files, no code files.
14. Scoring rubric is read-only — Read scoring.md for reference but never modify it.
15. Seed list is user-owned — Read critical-projects.json but never write to it.
16. Commit mode is gated — Requires user to say "commit" explicitly. Dry-run is always the default.