Junhanliu-dev/espalier-engineering / skills/espalier-init
espalier-init
VerifiedAnalyze any existing codebase, discover its patterns and best practices, and generate an Espalier structure (rules, skills, wiki, pipeline) so AI coders produce code matching that project's language and conventions. Greenfield repos take the two-pass Decide-Then-Bind path (/espalier-map charts the conventions, init binds them); boilerplate repos run init normally, then map the product. Wires into Claude Code, Codex, and/or GitHub Copilot (.claude/, .agents/skills/ + .codex/, .github/).
templates/agents/modes/simplification.md
View on GitHub# Mode: simplification
Read by `harness-coder` and `harness-reviewer` when requirements.md carries `simplify_from:` (the prompt says `SIMPLIFICATION CHANGE:`) — a cut filed by `/espalier-simplify`. Two sections: the coder's, then the reviewer's — read yours. Named on that prompt line by the orchestrator;
not loaded otherwise. Verbatim procedure — the agent body keeps the heading
and a one-line pointer; nothing here is a budget.
## Simplification Changes: Retire the Whole Obligation
When requirements.md carries `simplify_from:` frontmatter (a cut filed by
`/espalier-simplify`), the task is a DELETION with a proof record, and the
failure modes differ from a feature's: a half-removed surface, a consumer the
record missed, or complexity that merely moved. Read the
`## Simplification Evidence` and `## Retired Surface` sections first, then:
1. **Cut the whole boundary, from the outside in.** For every Retired
Surface entry take the declaration; its registration / dispatch /
parsing / compatibility paths; the implementations, adapters, state,
caches, events, and cleanup behind it; its imports / exports / barrels
and generated inventories; its config keys, env vars, and flags; its
migrations, fixtures, examples, and doc sections; the tests dedicated to
it; and any dependency or script that becomes unnecessary. Leave no
stub, no `removed` comment, no dead re-export, no compat shim, and no
"just in case" branch — a shim or migration path exists ONLY when the
requirement names the consumer that needs it.
2. **Never relocate the complexity.** Two representations do not become one
by adding a synchronization layer; a deleted wrapper does not come back
as a helper two files over. If the cut needs machinery the record did
not budget (`Net effect`), stop and report it — that is a survey error,
not a coding problem.
3. **A consumer the record missed is a STOP, not a workaround.** If a
search or a build / test failure reveals a live consumer — dynamic
dispatch by string, a template, a config key, a scheduled job, a
published export, persisted data — do not improvise a fallback or a
partial delete. Report it under `- Missed consumer:` with `file:line`,
leave the boundary consistent (fully cut or fully intact), and let the
round FAIL back to the survey. The same holds on a `FIX ROUND` carrying
a `[simplify-consumer]` / `[simplify-protected]` P0: either prove that
consumer dead too and widen `## Retired Surface` under the same
discipline, or restore the boundary fully — never patch a partial
delete; the orchestrator aborts the change and the survey row becomes
a LEAD carrying the fact.
4. **Run the residue search after the cut:** every retired name, string,
path, key, and flag, repo-wide (`grep -rn`), excluding git history and
CHANGELOG. Zero hits, or each remaining hit named with its reason.
5. **Keep the surviving contract proven.** Run the record's decisive check
and the surviving contract's tests in the scoped test command (folded
test-mode: they are this change's tests — add the one that would fail
if the cut were wrong when none exists). Never weaken or delete a test
to make the deletion pass; a test that only described the retired
behavior is part of the cut and is listed as such.
Append to your coding report:
```
### Retired Surface
- Retired: {every entry from requirements.md — each marked done}
- Residue search: {command} — {0 hits | N hits: file:line — reason each}
- Surviving contract check: {test / probe run — result}
- Kept on purpose: {entry — the consumer that needs it | none}
- Missed consumer: {file:line — what reaches it | none}
```
A simplification change whose report lacks this block, or whose
`Missed consumer:` line is not `none`, is not complete — the reviewer files
it and the cut returns to the survey page.
## Simplification Review (when the change retires surface)
Run this after step 5 and before the Minimalism Review whenever the coding
report carries a `### Retired Surface` block or requirements.md carries
`simplify_from:` frontmatter — the change is a proved deletion filed by
`/espalier-simplify`, and the panel is the second, independent proof. Read
`## Retired Surface` in requirements.md, then:
1. **Re-search every retired name yourself, BEFORE reading the coder's
residue search.** For each identifier, file, route, config key, env var,
flag, event / protocol string, table / column, and fixture: search
production code, tests, dynamic surfaces (string-keyed dispatch,
reflection, registries, templates, config and env files, scheduled jobs,
migrations, generated inventories), and external surfaces (published
package exports, API routes, persisted formats, downstream consumers
named in `espalier/wiki/external-services.md`). A live consumer the cut
broke or left dangling is a **P0 `[simplify-consumer]`** with the exact
`file:line` — the survey record was wrong, and the round FAILs however
clean the diff looks. On the fix round that follows, only two outcomes
pass: the boundary WIDENED with the consumer proved dead (re-search it
yourself), or the boundary fully RESTORED; a partial delete that merely
silences your `file:line` is the same P0 again.
2. **Boundary completeness.** A stub, a `removed` comment, a dead
re-export, an orphaned fixture / test / doc section / config key, or a
compat branch nothing reaches is a **P1 `[simplify-residue]`** naming
the leftover.
3. **Relocation, not removal.** New synchronization glue, a re-grown
helper, an adapter that preserves the retired policy — complexity moved
rather than retired — is a **P1 `[simplify-relocate]`** when it
contradicts the record's `Net effect`; a smaller instance is a `delete:`
Minimalism note.
4. **Protected surface.** An authorization or ownership check, input-trust
validation, data-loss guard, stored-format or migration compatibility
path, or cleanup that establishes quiescence that the diff retires
WITHOUT the requirement naming that consequence in its own words is a
**P0 `[simplify-protected]`** — the security auditor files its own; you
file yours.
5. **Proof present.** The record's decisive check and the surviving
contract's tests are in the diff's test scope and pass; a deleted or
weakened test on SURVIVING behavior is a **P1 `[simplify-proof]`** (a
test that only described the retired behavior is part of the cut). A
missing `### Retired Surface` block, or a `Missed consumer:` line other
than `none`, is a P1 — the change is not complete.
Lead the Problem cell with the tag (`[simplify-consumer] …`): the round
snapshot keeps only the first 80 characters and `espalier-stats.sh` counts
these tags.
Independent means independent: run your own searches first and compare with
the coder's residue search afterwards; a search the coder ran that you cannot
reproduce is itself a finding. Findings use the normal table, count in the
sentinel like any P0 / P1, and the tie-break holds — nothing mandated by the
rules or specs is ever residue.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |