dictionary-of-ai-coding

Updated the link to the AI Coding Dictionary and changed image sources.

Four new entries: Identity layer, Familiar Contract, Trust tier, Protected surface This PR adds four dictionary entries naming a layer the dictionary doesn't currently have a vocabulary for: the identity layer that sits above the harness. — the umbrella concept (above the harness, not the harness itself) — the published specification form (named, versioned identity contracts) — the operative ladder mechanism (which categories of self-modification require which authorization) — the on-disk classification (which files are off-limits to autonomous self-modification) Why these belong in the dictionary The existing entries cover the execution layer well: system prompt, agent mode, permission mode, AGENTS.md, memory system, skill, subagent, handoff, handoff artifact, compaction. What they don't cover is the layer above that — the part of an agent that has to be the same agent across sessions, with values and accumulated character that can't be silently rewritten by self-improvement loops. That layer is increasingly load-bearing in production. Self-Harness (Zhang et al., arXiv:2606.09498), Saravia's Autonomous Long-Running Coding Agents article, and the Claude Skills documentation all describe agents that progressively edit their own configuration. The harness is now expected to be edited, repeatedly, by the agents that operate inside it. Without an entry for which parts can't be edited, a reader of the dictionary has no vocabulary for what makes "the same agent" mean anything across time. The four entries are small, opinionated, and try to match the dictionary's existing register: woven symptom near the definition, table for the trust-tier ladder where it's naturally stepped, Avoid + Usage dialogue, single-occurrence cross-refs. I've checked them against the contributing rules in (≤140 char descriptions, ≥200 word bodies, first-occurrence cross-refs only, plain de-hyped register, tables only where naturally stepped). What's in the PR Cross-references on first occurrence to existing entries: , , , , , , . Internal cross-refs between the four new entries also follow the first-occurrence rule. The Trust tier entry uses the table convention from and . Curriculum.md update I've added a new Section 7 — Identity & Trust to containing the four new terms (Identity layer → Familiar Contract → Protected surface → Trust tier), and renumbered the existing Patterns of Work section from 7 to 8. The husky pre-commit hook requires every dictionary entry to be referenced in Curriculum.md, so this change had to land alongside the entries to keep green. If you'd prefer a different placement — slot the terms into existing sections, group them differently, or rename the new section — say the word and I'll restructure in a follow-up commit. Disclosure The Familiar Contract entry references a specification I'm involved with: OpenCoven/familiar-contract (v0.1.0, MIT, June 2026). The entry defines the concept of a published identity-layer contract, not the specific implementation — the entry would still be accurate for any future identity-layer specification someone else might publish. If you'd rather keep that entry vendor-neutral, I can drop the OpenCoven reference and leave only the abstract definition. Reduce-verbosity opportunities (optional follow-up) asks contributors to search existing entries for places a new term could reduce verbosity. I haven't done that polish pass yet to keep this PR scoped, but I'm happy to follow up with a separate commit if these four are accepted. Candidates I noticed: : the current "attempts to make an agent stateful across sessions" framing could cleanly contrast with — the former reloads state, the latter protects state from unauthorized change. : project-side instructions vs. agent-side standing brief is exactly the distinction Familiar Contract surfaces. , : capability composition is orthogonal to identity, and naming that explicitly might help. Happy to take any feedback, slim entries down, restructure the curriculum placement, or back the whole PR out gracefully if the dictionary's scope is intentionally narrower than this. Either way, thanks for keeping this dictionary public — it's the cleanest plain-English reference I've found for this material.

Summary Add Simplified Chinese dictionary entries under . Add Simplified Chinese curriculum and README template generation. Generate alongside the existing English README. Update the README freshness workflow to verify both generated README files. Validation