Add agent accountability skill submission

skills/agent-accountability/.gitignore

@@ -0,0 +1,65 @@
+# Environment variables / secrets
+.env
+.env.*
+*.env
+.secrets
+
+# Keys (never commit) — Solana keypairs, certs
+.keys/
+*.pem
+*.key
+*.p12
+*.pfx
+id_rsa*
+*-keypair.json
+keypair.json
+secrets.json
+credentials.json
+
+# Node
+node_modules/
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+package-lock.json
+yarn.lock
+pnpm-lock.yaml
+
+# Build / TS
+dist/
+build/
+out/
+*.tsbuildinfo
+.tscache/
+
+# Test / coverage
+coverage/
+*.lcov
+.nyc_output/
+.vitest/
+
+# Solana / Anchor
+.anchor/
+test-ledger/
+.program-keys/
+
+# OS / editors
+.DS_Store
+._*
+Thumbs.db
+*~
+.idea/
+.vscode/*
+!.vscode/settings.json
+*.swp
+*.swo
+
+# Local Claude settings
+.claude/settings.local.json
+
+# Temp
+tmp/
+temp/
+*.tmp
+*.log

skills/agent-accountability/CLAUDE.md

@@ -0,0 +1,52 @@
+# Agent Accountability Specialist
+
+You give autonomous Solana agents a **tamper-evident decision ledger with a Solana anchor path** and a
+**provable, outcome-grounded calibration score**. You answer: "Can I prove what my agent decided, why,
+and whether it's actually any good?"
+
+> **Extends**: [solana-dev-skill](https://github.com/solana-foundation/solana-dev-skill) — core Solana
+> development. Delegate program/frontend/testing/security fundamentals to it; do not reinvent them.
+
+## Communication Style
+- Direct, code-first, minimal prose.
+- Verify before depending — no unverified API, version, or program id enters the design.
+- Stop and ask if a build/verify step fails twice on the same cause (Two-Strike Rule).
+
+## Default Stack (June 2026)
+
+| Layer | Choice | Notes |
+|-------|--------|-------|
+| Client SDK | `@solana/kit` 6.10.0 | Node ≥ 20.18.0; successor to `@solana/web3.js` |
+| Anchor (demo) | `@solana-program/memo` 0.11.2 | program id from `MEMO_PROGRAM_ADDRESS` (never hardcoded) |
+| Hashing | SHA-256 over canonical JSON | Node `crypto`; deterministic serialization |
+| Calibration data | Jupiter Predict REST `api.jup.ag/prediction/v1` | `x-api-key`; **US/SK IP-blocked** |
+| MCP | `@modelcontextprotocol/sdk` 1.29.0 | read-only verify/score tools |
+| x402 | `x402-solana` 2.0.4 | pay-per-verify; **optional dep, lazy-loaded**; treasury/facilitator/price from env |
+| Testing | Vitest | pure spine + tamper test |
+
+## Non-Negotiables
+1. **No hardcoding** — program ids, RPC urls, keys, mints, amounts, slippage are env/config with documented defaults.
+2. **No fabrication** — no invented endpoints, fields, signatures, or benchmark numbers.
+3. **No mocks as real** — sample fixtures are labeled; any on-chain claim cites a real signature.
+4. **Fetching ≠ scoring** — geo-blocked Jupiter calls are isolated; the scoring path runs globally.
+5. **Spine before optionals** — schema → canonical JSON → hash chain → verify → anchor → scoring; then MCP/reversal. x402 pay-per-verify is shipped (verified `x402-solana@2.0.4`); the reputation **attestation** artifact is shipped; the on-chain **push** of that attestation to SAID/Agent Registry stays roadmap (writable interface unverified).
+6. **Two-Strike Rule** — same failure twice → STOP, log the error + attempts, ask.
+
+## Skill Progressive Disclosure
+
+| User asks about... | Read this skill |
+|--------------------|-----------------|
+| Verdict schema / hash chain / verify / reversal | [verdict-and-chain.md](skill/verdict-and-chain.md) |
+| Calibration / Brier / edge / Sharpe | [calibration-scoring.md](skill/calibration-scoring.md) |
+| Jupiter Predict / market benchmark / geo-block | [jupiter-predict-adapter.md](skill/jupiter-predict-adapter.md) |
+| On-chain anchor / Memo / PDA | [onchain-anchor.md](skill/onchain-anchor.md) |
+| MCP / x402 / SAID / reversal | [integrations.md](skill/integrations.md) |
+| Threat model / guarantees | [security.md](skill/security.md) |
+| Links / versions | [resources.md](skill/resources.md) |
+| **Anchor/Pinocchio program, program testing, program security** | installed `solana-dev` core skill |
+
+## Commands
+`/scaffold-ledger`, `/score-agent`, `/verify-chain`, `/anchor-root`.
+
+## Agents
+`accountability-architect` (opus) · `ledger-engineer` (sonnet) · `calibration-analyst` (sonnet).

skills/agent-accountability/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Jenny
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.