{"id":51195021,"url":"https://github.com/alexei-led/coding-posture","last_synced_at":"2026-06-30T22:00:36.564Z","repository":{"id":367027191,"uuid":"1278915857","full_name":"alexei-led/coding-posture","owner":"alexei-led","description":"A single SKILL.md skill giving coding agents task-aware working modes (debug, fix, review, migrate, …) — works with Hermes, Claude Code, Codex, Cursor, and Pi","archived":false,"fork":false,"pushed_at":"2026-06-25T06:55:40.000Z","size":81,"stargazers_count":11,"open_issues_count":0,"forks_count":1,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-29T21:27:46.919Z","etag":null,"topics":["agent-skills","ai-agents","claude-code","codex","coding-agents","cursor","hermes-agent","llm","pi","prompt-engineering","skill","skill-md"],"latest_commit_sha":null,"homepage":null,"language":"Shell","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/alexei-led.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":"AGENTS.md","dco":null,"cla":null}},"created_at":"2026-06-24T07:53:13.000Z","updated_at":"2026-06-29T18:07:04.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/alexei-led/coding-posture","commit_stats":null,"previous_names":["alexei-led/coding-posture-engine","alexei-led/coding-posture"],"tags_count":10,"template":false,"template_full_name":null,"purl":"pkg:github/alexei-led/coding-posture","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/alexei-led%2Fcoding-posture","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/alexei-led%2Fcoding-posture/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/alexei-led%2Fcoding-posture/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/alexei-led%2Fcoding-posture/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/alexei-led","download_url":"https://codeload.github.com/alexei-led/coding-posture/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/alexei-led%2Fcoding-posture/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34984784,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-30T02:00:05.919Z","response_time":92,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent-skills","ai-agents","claude-code","codex","coding-agents","cursor","hermes-agent","llm","pi","prompt-engineering","skill","skill-md"],"created_at":"2026-06-27T19:04:49.911Z","updated_at":"2026-06-30T22:00:36.544Z","avatar_url":"https://github.com/alexei-led.png","language":"Shell","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Coding Posture\n\nTask-aware working modes for coding agents. One `SKILL.md` file: before non-trivial work, the agent picks a mode — `debug`, `fix`, `review`, `test-first`, `refactor`, `optimize`, `migrate`, `upgrade`, `integrate`, `spike`, `unstuck` — and follows its checklist.\n\nIt exists to stop agents from behaving like optimistic elevators with write access: thrashing on a stuck bug, faking green tests, skipping reproduction, running destructive commands, migrating without a rollback.\n\n**Works with** Hermes · Claude Code · Codex · Cursor · Pi — any [`SKILL.md`](https://hermes-agent.nousresearch.com/docs/user-guide/features/skills)-compatible agent.\n\n## How the agent uses it\n\n```mermaid\nflowchart TD\n    Task[\"Non-trivial coding task\"] --\u003e Pick{\"Pick a mode that fits\u003cbr/\u003e(debug, fix, review, migrate, …)\"}\n    Pick --\u003e|no match| Normal[\"Proceed normally\"]\n    Pick --\u003e|match| Check[\"Follow the mode's checklist\"]\n    Check --\u003e Loop[\"gather context → localize → smallest change\u003cbr/\u003e→ run the real check → read output\"]\n    Loop --\u003e V{\"Verified by a real run?\"}\n    V --\u003e|no| Loop\n    V --\u003e|yes| Done([\"Done\"])\n    Guard[\"Always-on invariants:\u003cbr/\u003eno fake green · no destructive commands · don't game tests\"] -.-\u003e|guards every step| Check\n    Guard -.-\u003e Loop\n```\n\n## Modes\n\n| Mode         | Use when                                 | Core discipline                                     |\n| ------------ | ---------------------------------------- | --------------------------------------------------- |\n| `debug`      | failing test, bug, regression            | reproduce first, one hypothesis at a time           |\n| `fix`        | small known urgent change                | smallest diff, no opportunistic cleanup             |\n| `review`     | security/auth/payments, reviewing a diff | no approval without file/line evidence              |\n| `test-first` | behavior change, tests practical         | see RED before implementing, never fake green       |\n| `refactor`   | cleanup, simplify, rename                | preserve behavior, trace call sites before deleting |\n| `optimize`   | performance work, hot path               | measure first, baseline before/after                |\n| `migrate`    | schema/data/infra change                 | rollback path before touching state                 |\n| `upgrade`    | dependency or version bump               | read breaking changes, no blind search-replace      |\n| `integrate`  | calling an external API/service          | read the contract, handle the error paths           |\n| `spike`      | prototype, PoC, unknown library          | isolate, end with a verdict                         |\n| `unstuck`    | repeated failures, thrashing             | stop editing, summarize evidence, narrow hypotheses |\n\n**Always — in every mode:**\n\n- Verify by running the real check (test, build, repro); never by re-reading.\n- Never report a result you didn't run; never weaken, delete, or game a test to go green.\n- No destructive commands (`force push`, `reset --hard`, `drop`, `rm -rf`) without explicit scope.\n\n## Install\n\nInstall as a **plugin** (one command, updates in place) or drop the skill in directly. It's a standard `SKILL.md` skill, so it works unmodified across compatible agents.\n\n### Claude Code — plugin\n\n```text\n/plugin marketplace add alexei-led/coding-posture\n/plugin install coding-posture@coding-posture\n```\n\nUpdate later with `/plugin marketplace update coding-posture` then `/reload-plugins`, or enable auto-update from the `/plugin` menu to refresh on startup.\n\n### Codex CLI — plugin\n\n```bash\ngit clone git@github.com:alexei-led/coding-posture.git\ncodex   # then: /plugins → add the local marketplace in ./coding-posture → install coding-posture\n```\n\nOr point Codex straight at the skill in `~/.codex/config.json`:\n\n```json\n{ \"skills\": [\"/abs/path/to/coding-posture/skills/coding-posture\"] }\n```\n\nUpdate later with `git pull`.\n\n### Other agents\n\n| Agent                             | Install                                                                                                                                                                                      |\n| --------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| **Hermes**                        | Drop `skills/coding-posture/` into `~/.hermes/skills/` (auto-discovered). URL/hub install: see the [Hermes skills docs](https://hermes-agent.nousresearch.com/docs/guides/work-with-skills). |\n| **Pi**                            | `pi install git:github.com/alexei-led/coding-posture`                                                                                                                                        |\n| **Cursor / any `SKILL.md` agent** | Copy `skills/coding-posture/` into the agent's skills dir.                                                                                                                                   |\n\nThe agent activates the skill from its `description` when a coding task starts.\n\n## How it works — theory and evidence\n\nA skill is just text placed in the model's context. A worked procedure acts as an _in-context demonstration_: the model conditions its next tokens on the shown trajectory, not just the final answer — the same basis as chain-of-thought. No complete mechanistic theory of frontier-model reasoning exists yet, so treat this as a grounded substrate, not a proof.\n\nEach design claim, and how strong the evidence actually is:\n\n| Claim                                                   | Evidence strength                             | Sources                                                                                                                                                                                                                                                                                                                                                                                      |\n| ------------------------------------------------------- | --------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |\n| A checklist shifts behavior via in-context conditioning | Grounded substrate, not a full theory         | [CoT (Wei 2022)](https://arxiv.org/abs/2201.11903), [induction heads (Olsson 2022)](https://transformer-circuits.pub/2022/in-context-learning-and-induction-heads/index.html), [iteration heads (NeurIPS 2024)](https://proceedings.neurips.cc/paper_files/paper/2024/file/c50f8180ef34060ec59b75d6e1220f7a-Paper-Conference.pdf), [ICL/CoT (ICML 2024)](https://icml.cc/virtual/2024/38391) |\n| Procedures beat personas                                | Supported guideline, not a law                | [personas don't help (Zheng 2024)](https://aclanthology.org/2024.findings-emnlp.888/), [self-consistency (2022)](https://arxiv.org/abs/2203.11171)                                                                                                                                                                                                                                           |\n| The model self-selects the mode                         | Holds for strong models, not universal        | [Route-to-Reason (2025)](https://arxiv.org/abs/2505.19435)                                                                                                                                                                                                                                                                                                                                   |\n| The checklists target the right levers                  | Highest-evidence levers + documented failures | [self-debug (Chen 2023)](https://arxiv.org/abs/2304.05128), [self-correction limits (Huang 2023)](https://arxiv.org/abs/2310.01798), [context-first (2026)](https://arxiv.org/abs/2604.02547), [verifier gaming (2026)](https://arxiv.org/abs/2604.15149)                                                                                                                                    |\n| The skill helps                                         | +15pp in the eval — run it yourself           | [`eval/`](eval/)                                                                                                                                                                                                                                                                                                                                                                             |\n\nTwo deliberate consequences:\n\n- **Procedures, not personas.** A persona (\"act as an expert debugger\") sets style; a procedure supplies structure the model can follow. So each mode is a checklist, not a character.\n- **Small on purpose.** Instruction-following degrades as a prompt grows long and complex, so a short, followable procedure beats a long aspirational one. (The popular \"~150–200 instructions\" ceiling has no peer-reviewed source; the real effect is just degradation with length.)\n\n## Where this fits\n\nFrontier agents already do the basics (reproduce bugs, run tests, small diffs), and a good `CLAUDE.md`/`AGENTS.md` repeats them. This skill earns its place only as the **delta** — the anti-instincts agents get wrong by default (stop thrashing, don't game the grader, roll back before migrating, measure before optimizing, read the API contract) — plus a mode catalog too large to keep always-on.\n\nSplit the rules by how often they must fire:\n\n| Layer                                     | What goes here                                                  | Why                                                         |\n| ----------------------------------------- | --------------------------------------------------------------- | ----------------------------------------------------------- |\n| **Always-on** — `CLAUDE.md` / `AGENTS.md` | the invariants ([`always-on-snippet.md`](always-on-snippet.md)) | must fire every turn; a conditional skill can miss-activate |\n| **Conditional** — this skill              | the per-task mode checklists                                    | loaded only when relevant; too many to keep always-on       |\n\n[Anthropic's skill guidance](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices) is evaluation-driven: keep only what measurably closes a gap.\n\n## Try it\n\nOpen source — clone it, drop the skill into your agent, and watch how it changes the work. In the behavioral eval ([`eval/`](eval/)) the skill scores **85% vs 70% without it (+15pp)**, with the urgent-auth case going 4/4 vs 2/4. Run the eval on your own model and tasks, [open an issue](https://github.com/alexei-led/coding-posture/issues) with what you find, or add a mode — contributions welcome.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Falexei-led%2Fcoding-posture","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Falexei-led%2Fcoding-posture","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Falexei-led%2Fcoding-posture/lists"}