smith-guidance

github:tianjianjiang/smith · first committed 2026-01-10 · source · badge

Core agent steering with HHH framework (Helpful, Honest, Harmless), exploration-before-implementation workflow, and anti-sycophancy rules. Use when guiding AI agent behavior, handling disagreements, or establishing interaction patterns. Always active for all agent interactions.

Skill Health: not yet scored

Listed, not yet scored. This skill has no public model test yet. It is in the scoring queue; submitted skills go first, then the most-installed.

Version history

37 commits to this file on GitHub (newest first).

2026-10-20 · commit 6e431f3 · docs(guidance): forbid anthropomorphic self-reference (#321)
2026-10-17 · commit bee37de · fix(guidance): restore the verbatim deviation report to the always-loaded rule (#308)
2026-10-17 · commit d325571 · docs(guidance): require full conditions when a decision is asked or recorded (#307)
2026-10-17 · commit d1ff329 · fix(ctx-claude): require a chat preview and the user's reply before an external write (#306)
2026-10-16 · commit ca7d3fb · docs(guidance): base the consent rule on a transaction model (#304)
2026-10-15 · commit a3e85af · docs(guidance): compress smith-guidance under the 2,000-token budget (#299)
2026-10-14 · commit 463b756 · docs(guidance): route general corrections to a general place (#297)
2026-10-05 · commit aa6cc5c · docs: remove dated and conflicting prompt text (#292)
2026-10-02 · commit 4ffb46c · feat(skills): say when to use each skill in its description (#290)
2026-10-02 · commit 438aeb1 · feat(smith-ctx-claude): lint skill files for point-in-time content (#288)
2026-09-11 · commit 91d9dd7 · fix(gh-pr): split bot vs human reviewer in review-cycle reply/resolve (#272)
2026-09-03 · commit 3a3e842 · refactor(smith-guidance): remove no-op HHH definitions (#244)
2026-08-30 · commit 943d76e · refactor(skills): compress frontmatter descriptions for token efficiency (#209)
2026-08-30 · commit 2586277 · refactor(skills): deduplicate HHH framework between smith-principles and smith-guidance (#207)
2026-08-29 · commit cbf081f · fix: remove auto-memory references after feature disabled (#193)
2026-08-22 · commit e5f7e9d · docs(guidance): require a decision request to carry its own context (#169)
2026-08-13 · commit 55bd0a2 · refactor(skills): dissolve smith-stack and smith-stacks into smith-ship and smith-gh-pr (#164)
2026-08-13 · commit d3ff97d · refactor(skills): fold recorded-decision discipline into smith-validation and smith-guidance (#163)
2026-07-28 · commit d3b56c8 · feat(validation): make adversarial verification of findings canonical (#157)
2026-07-28 · commit 8ae4cdc · docs(skills): record how a review can look clean without running (#155)
2026-07-17 · commit b440689 · refactor(skills): single-source the external-write gate (#144)
2026-07-12 · commit f856231 · docs(skills): drop XML tag skeleton for plain Markdown across all skills (#128)
2026-07-11 · commit 25fe82a · docs(guidance): add scoped-edit and goal-setting rules (#127)
2026-07-11 · commit 9905f31 · feat(hooks): enforce branch-first edits and dirty-tree worktree guard (#125)
2026-06-30 · commit 846c398 · docs: harden ship-review and external-write discipline across skills (#119)
2026-06-27 · commit 0f6199c · docs(skills): trim always-loaded surface ~27%, move catalog+refs to REFERENCE (#117)
2026-06-26 · commit 051f2a5 · docs(guidance): require closing gaps, not just disclosing them (#112)
2026-06-25 · commit 9973039 · docs(skills): make repeatedly-forgotten rules reachable at decision time (#107)
2026-06-25 · commit 579db24 · feat(smith-subagents): new skill — spawn narrow, trust returns nothing (#104)
2026-06-25 · commit 44efe1a · docs(cc-env wave2): process rules — output-to-file, verify-from-source, sweep/zsh, ff-only pull (#103)
2026-06-25 · commit 418d76b · docs(smith-skills): wave-2 over-budget additions (serena, guidance, tools) (#98)
2026-06-15 · commit 22e6ff5 · feat(guidance): require verifying external-dependency mechanisms before proposing (#89)
2026-05-24 · commit 7a7c35a · feat(skills): generalize 23 lessons learned into 6 skill sections (#83)
2026-03-10 · commit d6fc69d · fix(guidance): enhance commit guardrails and PR review standards (#61)
2026-03-01 · commit d98c949 · feat: add pre-operation verification and proactive testing rules (#58)
2026-02-07 · commit 5ef711f · feat: add guardrails from insights friction analysis (#51)
2026-01-10 · commit 81c7c35 · refactor: rename all skills with smith- prefix for Claude Code compatibility (#50)

Archived snapshots