Some checks are pending
Mirror PR to Forgejo / mirror (pull_request) Waiting to run
- What: New submit skill (PR mechanics for all agents) + GitHub Actions workflow that mirrors external contributor PRs to Forgejo for eval - Why: Agents need a single reference for the Forgejo PR workflow. External contributors need a path from GitHub to our eval pipeline. - submit.md complements extract.md: extract = how to produce claims, submit = how to get them into the knowledge base Pentagon-Agent: Leo <B9E87C91-8D2A-42C0-AA43-4874B1A67642> Model: claude-opus-4-6
195 lines
6 KiB
Markdown
195 lines
6 KiB
Markdown
# Skill: Submit
|
|
|
|
Get your claims and source archives into the knowledge base via PR.
|
|
|
|
## When to Use
|
|
|
|
After running `skills/extract.md` — you have claim files and source archives ready to propose.
|
|
|
|
## Prerequisites
|
|
|
|
Your Forgejo token is at `~/.pentagon/secrets/forgejo-{your-name}-token` (e.g., `forgejo-rio-token`).
|
|
|
|
Set up the git remote once per worktree:
|
|
|
|
```bash
|
|
AGENT_TOKEN=$(cat ~/.pentagon/secrets/forgejo-{your-name}-token)
|
|
git remote add forgejo https://{your-name}:${AGENT_TOKEN}@git.livingip.xyz/teleo/teleo-codex.git
|
|
```
|
|
|
|
If the `forgejo` remote already exists, skip this. You can check with `git remote -v`.
|
|
|
|
## Process
|
|
|
|
### Step 1: Branch from latest main
|
|
|
|
```bash
|
|
git fetch forgejo
|
|
git checkout -b {your-name}/{brief-description} forgejo/main
|
|
```
|
|
|
|
Branch names: `rio/displacement-claims`, `clay/shapiro-extraction`, `theseus/fep-batch-1`, etc.
|
|
|
|
### Step 2: Create source archives
|
|
|
|
For each source, create a file in `inbox/archive/` following `schemas/source.md`:
|
|
|
|
```yaml
|
|
---
|
|
type: source
|
|
title: "Article title"
|
|
author: "Name (@handle)"
|
|
twitter_id: "stable numeric ID if from X"
|
|
url: https://example.com/article
|
|
date: 2026-03-09
|
|
domain: internet-finance
|
|
format: essay
|
|
status: unprocessed
|
|
tags: [topic1, topic2]
|
|
---
|
|
```
|
|
|
|
If the source is from X, always include `twitter_id` — handles change, IDs don't. Get the ID from the tweet author object or via `/x-research`.
|
|
|
|
### Step 3: Write claim files
|
|
|
|
Create `.md` files in `domains/{your-domain}/` with proper YAML frontmatter:
|
|
|
|
```yaml
|
|
---
|
|
type: claim
|
|
domain: internet-finance
|
|
description: "one sentence adding context beyond the title"
|
|
confidence: proven | likely | experimental | speculative
|
|
source: "who proposed this and primary evidence"
|
|
created: 2026-03-09
|
|
---
|
|
```
|
|
|
|
- One claim per file
|
|
- Filename = slugified title (lowercase, hyphens, no special chars)
|
|
- Title IS the claim — prose proposition, not a label
|
|
- Evidence cited inline in the body
|
|
- Wiki links `[[to related claims]]` where they exist
|
|
|
|
See CLAUDE.md "Claim Schema" for full spec.
|
|
|
|
### Step 4: Update source archive status
|
|
|
|
After extraction, update the source file frontmatter:
|
|
|
|
```yaml
|
|
status: processed
|
|
processed_by: {your-name}
|
|
processed_date: 2026-03-09
|
|
claims_extracted:
|
|
- "claim title 1"
|
|
- "claim title 2"
|
|
enrichments:
|
|
- "existing claim that was updated"
|
|
```
|
|
|
|
### Step 5: Commit with trailers
|
|
|
|
```bash
|
|
git add domains/{your-domain}/*.md inbox/archive/*.md
|
|
git commit -m "{your-name}: add N claims about {topic}
|
|
|
|
- What: brief description of claims added
|
|
- Why: source material reference, why these matter
|
|
- Connections: what existing claims these relate to
|
|
|
|
Pentagon-Agent: {YourName} <{your-pentagon-UUID}>
|
|
Model: {your-model-id}"
|
|
```
|
|
|
|
Both trailers are required on every commit. Find your Pentagon UUID in your agent config. Model ID is the exact model you're running on (e.g., `claude-opus-4-6`, `claude-sonnet-4-5-20250514`).
|
|
|
|
### Step 6: Push to Forgejo
|
|
|
|
```bash
|
|
git push forgejo {your-name}/{brief-description}
|
|
```
|
|
|
|
**Push to the `forgejo` remote only. Never push to `origin` (GitHub).** GitHub is a read-only mirror. Pushing to GitHub will cause your branch to be deleted and your PR auto-closed by the mirror sync.
|
|
|
|
### Step 7: Create PR via Forgejo API
|
|
|
|
```bash
|
|
AGENT_TOKEN=$(cat ~/.pentagon/secrets/forgejo-{your-name}-token)
|
|
|
|
curl -s -X POST \
|
|
-H "Authorization: token $AGENT_TOKEN" \
|
|
-H "Content-Type: application/json" \
|
|
-d "$(jq -n \
|
|
--arg title '{your-name}: brief PR title' \
|
|
--arg body 'Summary of claims proposed.
|
|
Source: [reference]
|
|
Why: [what these add to the knowledge base]
|
|
Connections: [existing claims these relate to or challenge]' \
|
|
--arg head '{your-name}/{brief-description}' \
|
|
'{title: $title, body: $body, head: $head, base: "main"}')" \
|
|
"https://git.livingip.xyz/api/v1/repos/teleo/teleo-codex/pulls"
|
|
```
|
|
|
|
The PR body should include: summary of claims, source reference, why they add value, and any claims that challenge or extend existing ones.
|
|
|
|
### Step 8: Wait for review
|
|
|
|
The eval pipeline runs automatically every 2 minutes:
|
|
|
|
1. **Leo** reviews (cross-domain quality, on opus)
|
|
2. **Domain peer** reviews (domain expertise, on sonnet)
|
|
3. **Self-review** of your own PR (adversarial, on alternate model)
|
|
|
|
Outcomes:
|
|
- **All approve** — auto-merge via squash
|
|
- **Changes requested** — read the review comments, fix on the same branch, push again. The pipeline re-evaluates automatically.
|
|
|
|
## Checking PR Status
|
|
|
|
```bash
|
|
AGENT_TOKEN=$(cat ~/.pentagon/secrets/forgejo-{your-name}-token)
|
|
|
|
# List your open PRs
|
|
curl -s -H "Authorization: token $AGENT_TOKEN" \
|
|
"https://git.livingip.xyz/api/v1/repos/teleo/teleo-codex/pulls?state=open" \
|
|
| jq '.[] | {number, title, state}'
|
|
|
|
# Read review comments on a specific PR
|
|
curl -s -H "Authorization: token $AGENT_TOKEN" \
|
|
"https://git.livingip.xyz/api/v1/repos/teleo/teleo-codex/pulls/{PR_NUMBER}/reviews" \
|
|
| jq '.[] | {user: .user.login, state: .state, body: .body}'
|
|
```
|
|
|
|
## Handling Review Feedback
|
|
|
|
When a reviewer requests changes:
|
|
|
|
1. Read the review comments carefully
|
|
2. **Mechanical fixes** (broken wiki links, missing frontmatter, schema issues) — fix immediately
|
|
3. **Substantive feedback** (confidence calibration, reframing, domain classification) — exercise judgment, make changes you agree with
|
|
4. If you disagree with feedback, comment on the PR explaining your reasoning
|
|
5. Commit fixes to the same branch and push — the pipeline re-evaluates
|
|
|
|
```bash
|
|
git add -u
|
|
git commit -m "{your-name}: address review feedback on PR #{number}
|
|
|
|
- Fixed: [what you changed]
|
|
|
|
Pentagon-Agent: {YourName} <{your-UUID}>
|
|
Model: {model-id}"
|
|
git push forgejo {your-name}/{brief-description}
|
|
```
|
|
|
|
**Do not start new extraction work while you have PRs with requested changes.** Fix first, then move on.
|
|
|
|
## Rules
|
|
|
|
1. **Never push to GitHub.** Only push to the `forgejo` remote.
|
|
2. **Never commit to main.** Always branch + PR.
|
|
3. **Never merge your own PR.** The eval pipeline handles merge.
|
|
4. **Always include both git trailers** (Pentagon-Agent and Model).
|
|
5. **Always archive the source** before or alongside claim extraction.
|
|
6. **Always update source status** after extraction completes.
|