SEED
Definition: A fleshed-out concept, idea, or project in development — not yet ready for operational execution. Can also remain a SEED forever as a catalogued concept without needing to go anywhere.
Characteristics: - Has structure: problem statement, context, key questions, knowns/unknowns - Lives indefinitely as a catalogued artifact - Graduation is optional, not required - When it does graduate, the SEED doc persists and gets linked to the evolved entity
Lifecycle: SEED → (may graduate) → SPROUT (or remain SEED forever)
Relationship to SPARK: - A SPARK may “unfold” into a
SEED as a separate document - Both exist independently after the
unfolding - The SPARK gets tagged with
evolved_into: SEED-XXX - The SEED gets tagged with
evolved_from: SPARK-XXX
Example:
# SEED: Non-interactive Agent Access Architecture
**Filed:** 2026-05-13
**Status:** ACTIVE
**Type:** SEED
**Associations:**
- evolved_from: [SPARK-001]
-Evolves into: [SPROUT-001]
- Related: [POD-XXX]
**Problem:** Tailscale SSH intercept blocks SSH-based tooling for non-interactive
agent sessions. Remote operations that depend on ssh/scp/rsync fail.
**Context:** All six Forgejo repos affected. Agents cannot run remote commands
via ssh relik-pi4 <command>. Workaround uses Forgejo HTTP API but leaves
other remote tasks inaccessible.
**Key Questions:**
- What is the full scope of blocked operations?
- Is Tailscale SSH the only path or are there alternatives?
- What does a viable long-term access architecture look like?
**Known:**
- Tailscale web auth intercept is the blocker
- Forgejo API works as fallback for repo ops
- pi-starter-pack issue #8 tracks this
**Unknowns:**
- ACL configuration options
- Non-interactive key auth feasibility
**Tags:** #infra #access #tailnet #agents