contribute

name	contribute
description	Extract reusable knowledge from the current session, privacy-scrub it, and (with explicit user consent) submit it to the shared StatsClaw brain-seedbank as a pull request. Use this skill when the user says "contribute", "share what I learned", "submit lessons", "add to brain", "contribute knowledge", or invokes $contribute. Bypasses the automatic frequency heuristic but applies the same quality gate and mandatory user-consent step — nothing is uploaded without explicit approval.
metadata	{"short-description":"Contribute session knowledge to the shared brain"}

Skill: Contribute — Session Knowledge Contribution

The /contribute command lets users explicitly trigger knowledge extraction and contribution at any point during a session. It summarizes what worked, what required manual intervention, and what domain-specific patterns emerged — then offers to submit these lessons to the shared brain.

This is the user-invocable entry point for brain contributions. While the standard workflow dispatches distiller automatically after noteworthy workflows (brain-sync Phase 3), /contribute gives users direct control: they can invoke it whenever they want to share what they've learned.

---

Trigger Phrases

User says	Detected intent
/contribute	Direct skill invocation
"contribute" / "contribute to brain" / "share what I learned"	Natural language trigger
"submit lessons" / "share knowledge" / "add to brain"	Natural language trigger
"summarize lessons" / "what did we learn" + intent to share	Natural language trigger

---

When to Use

• After completing a workflow — the user wants to contribute knowledge from the current or recent session
• Mid-session — the user has accumulated insights across multiple runs and wants to batch-contribute
• Standalone — no active workflow required; the user may invoke /contribute purely to share knowledge from their experience

---

Prerequisites

1. Brain mode must be "connected". If brain mode is "isolated" or unset:

   • Leader asks the user via a numbered-options markdown question: "Brain mode is currently disabled. Would you like to enable it to contribute knowledge?"
   • If the user says yes: set BrainMode: "connected" in context.md, acquire brain repos (Phase 1 of skills/brain-sync/SKILL.md), then proceed.
   • If the user says no: inform them that /contribute requires brain mode and stop.

2. Brain repos must be available. If .repos/brain/ or .repos/brain-seedbank/ are not cloned:

   • Leader acquires them (same as brain-sync Phase 1).
   • If acquisition fails: warn the user and stop.

3. Workspace repo should be available (for storing the contribution artifact in the run directory). If not, /contribute still proceeds — the contribution goes to brain-seedbank regardless.

4. Push credentials for the user's fork of statsclaw/brain-seedbank must be verifiable. Leader checks during the flow and warns if push access cannot be confirmed.

---

Parameters

Parameter	Source	Default
scope	User prompt or inferred	"session" — all runs in the current session
run_id	User prompt (optional)	Latest completed run if not specified
target_repo	User prompt or inferred from context	Current active repo from context.md

---

Execution Flow

Step 1 — Gather Artifacts

Leader collects all available run artifacts for the specified scope:

Session scope (default): scan .repos/workspace/<repo-name>/runs/ for all runs from the current session. Collect from each run:

• comprehension.md, spec.md, test-spec.md, sim-spec.md
• implementation.md, simulation.md, audit.md
• log-entry.md, docs.md, mailbox.md
• review.md

Single run scope: collect artifacts from the specified <run_id> only.

If no artifacts are found (e.g., no runs have been executed yet), inform the user: "No workflow artifacts found to extract knowledge from. Run a workflow first, then use /contribute."

Step 2 — Create Contribution Run

Generate a contribution-specific run directory:

• Run ID format: CONTRIBUTE-<timestamp>
• Create .repos/workspace/<repo-name>/runs/CONTRIBUTE-<timestamp>/
• Write request.md with scope, source runs, and intent

Step 3 — Dispatch Distiller

Leader dispatches the distiller agent with:

• All collected artifact paths from Step 1
• Path to .repos/brain/index.md for duplicate checking
• Path to .repos/brain/ for existing entry browsing
• Instruction to produce brain-contributions.md in the contribution run directory

Distiller follows its standard workflow (agents/distiller.md):

1. Scan artifacts for extractable knowledge
2. Apply the 5-question quality gate
3. Apply privacy scrub (skills/privacy-scrub/SKILL.md)
4. Check for duplicates against brain index
5. Format entries using templates/brain-entry.md
6. Write brain-contributions.md

Step 4 — Present to User (Mandatory Consent)

Leader reads brain-contributions.md and presents the FULL content to the user via a numbered-options markdown question:

Here's a summary of what StatsClaw learned during this session:

---
[Full content of brain-contributions.md, showing each proposed entry with:
 - What worked well
 - What required manual intervention
 - What domain-specific patterns emerged]
---

Would you like to contribute this to StatsClaw's shared brain?
Your contribution will be publicly visible on GitHub (statsclaw/brain-seedbank)
with your GitHub username as the contributor.

1. Yes — contribute all entries above
2. Yes, but let me pick which ones (partial)
3. No — skip this time

This consent step is MANDATORY and NEVER skipped. The user always sees exactly what would be shared.

Handle responses:

• Yes (full): mark all entries as approved, proceed to Step 5
• Partial: ask which entries to keep, update brain-contributions.md to remove declined entries, proceed to Step 5
• No: inform the user that no contributions were made, clean up the contribution run directory, stop

Step 5 — Dispatch Shipper (Brain Upload Only)

Leader dispatches shipper with a brain-upload-only task:

• brain-contributions.md path
• User approval status
• No target repo push needed (this is brain-only)
• Workspace sync: copy brain-contributions.md and a minimal log to the workspace repo

Shipper follows its brain upload workflow (agents/shipper.md Step 7b):

1. Fork statsclaw/brain-seedbank (if not already forked)
2. Create contribution branch: contribute/<date>-<short-slug>
3. Write knowledge entry files to correct directories
4. Update index.md on the branch
5. Commit and push to user's fork
6. Create PR from fork to statsclaw/brain-seedbank

Step 6 — Report

Leader reports to the user:

• Number of entries contributed
• PR URL on brain-seedbank (if successful)
• Any entries that were rejected by the quality gate (with reasons)
• Reminder: "An admin will review your contribution. Accepted entries earn badges!"

---

Output Artifacts

Artifact	Location	Purpose
request.md	Contribution run directory	Records contribution scope and intent
brain-contributions.md	Contribution run directory	Proposed entries (from distiller)
shipper.md	Contribution run directory	Brain-seedbank PR URL and status

---

State Model

The /contribute command uses a simplified state flow (no full pipeline):

BRAIN_CHECK → ARTIFACTS_GATHERED → DISTILLER_DISPATCHED → USER_CONSENT → UPLOAD → DONE

State	Meaning
BRAIN_CHECK	Verifying brain mode and repo availability
ARTIFACTS_GATHERED	Collected run artifacts for extraction
DISTILLER_DISPATCHED	Distiller is extracting knowledge
USER_CONSENT	Waiting for user to approve/decline
UPLOAD	Shipper is creating brain-seedbank PR
DONE	Contribution complete (or declined)

---

Relationship to Standard Brain Workflow

Aspect	Standard (brain-sync Phase 3-5)	/contribute
Trigger	Automatic after noteworthy workflows	User-invoked at any time
Scope	Single run	Session-wide or single run
Frequency heuristic	Leader decides if distiller should run	Always runs (user explicitly asked)
Distiller	Same agent, same rules	Same agent, same rules
Privacy scrub	Mandatory	Mandatory
User consent	Mandatory	Mandatory
Upload	Via shipper (end of workflow)	Via shipper (standalone dispatch)
Quality gate	5-question gate	5-question gate

The key difference: /contribute bypasses the frequency heuristic because the user has explicitly signaled intent to share knowledge. Everything else (distiller, privacy scrub, quality gate, user consent, shipper upload) is identical.

---

Edge Cases

• No artifacts: If no runs exist in the current session, inform the user and stop.
• All entries rejected: If distiller's quality gate rejects all candidates, inform the user: "No entries met the quality threshold for contribution. This is normal for routine changes."
• Brain repos unavailable: Warn the user and stop. Do not proceed without brain repos.
• Already contributed: If the latest run already has a brain-contributions.md from the standard workflow, distiller should check for duplicates against it to avoid re-proposing the same entries.
• Mixed repos: If the user worked on multiple target repos in one session, /contribute extracts from all of them (scanning each repo's workspace runs).

---

Non-Blocking Guarantee

Like all brain features, /contribute failures are non-blocking:

• If distiller fails: warn the user, stop gracefully
• If shipper fails to create PR: warn the user, note that entries are saved locally in brain-contributions.md
• If brain repos can't be cloned: warn and stop
• The user's main workflow is never affected by /contribute failures

---

Example Usage

User: /contribute
→ Leader checks brain mode (connected? repos available?)
→ Leader gathers artifacts from all session runs
→ Leader dispatches distiller
→ Distiller extracts 3 entries: a math method, a coding pattern, a validation strategy
→ Leader shows all 3 entries to user
→ User: "Yes, contribute all"
→ Leader dispatches shipper for brain upload
→ Shipper creates PR to statsclaw/brain-seedbank
→ Leader: "3 entries contributed! PR: https://github.com/statsclaw/brain-seedbank/pull/42"

User: contribute what we learned about panel data
→ Same flow, but distiller focuses extraction on panel-data-related knowledge
→ User reviews, approves 2 of 3 entries
→ PR created with 2 entries

User: share the lessons from run REQ-20260404-001
→ Leader scopes to single run
→ Same extraction and consent flow