Skip to content

donaldfilimon/abi / .agents/skills/sea

Plan abi SEA (Sparse Evidence Attention) self-learning work — evidence-augmented completion, 8-signal scorer, EMA modulator persistence, and constitution audit. Use when working on ai_learn / complete --learn / evidence recall. Routes to sea-learn-loop, abi-superpower-sea, abi-superpower-constitution, and sea-learning-controller.

FileKindSize
SKILL.mdskill md1.7 KB
---
name: sea
description: Plan abi SEA (Sparse Evidence Attention) self-learning work — evidence-augmented completion, 8-signal scorer, EMA modulator persistence, and constitution audit. Use when working on ai_learn / complete --learn / evidence recall. Routes to sea-learn-loop, abi-superpower-sea, abi-superpower-constitution, and sea-learning-controller.
---

# sea

Entry point for abi's SEA self-learning loop (`crates/abi-sea/src/`). Routes:

| You want to… | Use |
| --- | --- |
| Drive `abi complete --learn` end-to-end | `sea-learn-loop` |
| Deep-dive SEA scoring / adaptive modulation | `abi-superpower-sea` |
| Constitution audit (observability-only) | `abi-superpower-constitution` |
| Run learning or inspect the REPL's session-only `/sea` preference | `sea-learning-controller` |

## Facts that constrain any SEA plan
- SEA = evidence-augmented self-learning completion with an 8-signal scorer +
  budgeted greedy selection. It classifies 7 task types; 3 currently apply
  non-baseline signal-weight deltas and 4 retain the baseline weights.
- `AdaptiveModulator` weights (EMA, `alpha=0.3`, key `modulator:weights`)
  persist in WDBX **only on the `--learn`/SEA path**. Plain `complete` re-runs
  sentiment each turn with no EMA persistence.
- Constitution audit (6 principles) is **observability-only, not a gate** —
  sets `audit_passed` / `audit_vetoed` / `escore`, warns via std.log on violation,
  but `complete` / `run` still return the response. Safety+privacy hard-veto
  only when either < 0.5.
- The live EMA path runs through `abi-sea::run_learn_loop`; there is no separate
  public adaptive-routing command.

## Honest boundary
No accuracy / energy / learning-gain claims without a repo benchmark.
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 196kUpdated Claude CodeCodexVerifiedInstall