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/wiki-sweep.md
View on GitHub# Mode: wiki-sweep
Read by `harness-reviewer` when the orchestrator's prompt line says
`WIKI SWEEP:` — the end of Stage 7, after the code is committed and before
`espalier-pr.sh project` builds the espalier branch. Named on that prompt line;
not loaded otherwise. Verbatim procedure — the agent body keeps the heading and
a one-line pointer; nothing here is a budget.
## What This Is Not
This is not `/espalier-prune`. Prune re-derives a doc by re-spawning the
discovery scouts against the whole codebase and gates every file on a human
accept. This sweep reads ONE diff — the change that just passed review — and
asks a single question per doc: **does this diff make anything written here
false?** It never re-derives, never re-scouts, and never rewrites a doc in its
own voice. Drift the sweep cannot see from this diff is the doctor's job.
Nothing here is your opinion about the docs. A wiki paragraph you find thin,
a rule you would have worded differently, a section you think is missing — all
`NO-IMPACT` unless THIS diff falsified it.
## Inputs
- The change's code diff: `git diff <Base-Ref>..HEAD -- . ':(exclude)espalier/'`
- `espalier/wiki/*.md` — architecture, data-models, critical-paths, external-services
- `espalier/rules/*.md`
## Procedure
1. **Skip fast, and say so.** If the diff touches no source file — tests only,
docs only, config only — every verdict is `NO-IMPACT` and you are done. Do
not go looking for something to justify the call.
2. **Read the diff before the docs.** Build the list of what actually changed:
modules added or removed, call paths rerouted, data shapes altered, external
calls introduced or dropped, entry points moved. Then read each doc against
that list. Reading the docs first invites you to pattern-match prose onto a
diff that does not support it.
3. **Per file, one verdict.** For each wiki and rule file: `NO-IMPACT`, or a
patch. A patch carries the smallest edit that makes the file true again —
amend the sentence, add the row, extend the list. Keep the file's existing
structure, headings, ordering and voice; a diff that reflows a section you
are not correcting is a rewrite, not a sweep.
4. **Cite or drop it.** Every patch names the hunk that forces it —
`file:line` from the DIFF, not from the doc. A claim you cannot anchor to a
hunk does not go in, however true it looks. This is the whole bar: the sweep
is trusted to commit without a human accept ONLY because every line it
writes is traceable to reviewed code.
5. **Never delete.** Additive or amending only. A section the diff appears to
retire is NOT removed — record it under `### Proposed Removals` and leave
the file's text alone. A removal needs the full prune protocol and its
Removed-rules ledger, because a silently dropped rule is unrecoverable and
invisible. Retiring surface is `/espalier-simplify`'s lane, not this one.
6. **Rules are canon, and the bar is higher.** Edit `espalier/rules/` ONLY when
the diff establishes a convention the rules do not yet state, or contradicts
one they do. A rule file is not a changelog: a change that merely FOLLOWS an
existing rule is `NO-IMPACT`. When you do patch a rule, say in the
justification which rule bullet is now wrong and why the diff proves it — a
rule edit routes the espalier PR to the rule owner (CODEOWNERS, Rule-canon
gate in `espalier/rules/development-process.md`), so a speculative one costs
a human their time.
7. **Silence is a correct answer, and the common one.** Most changes do not
move architecture, data models, critical paths or external services. A sweep
that finds something every run is miscalibrated, and the patches it writes
are noise in a PR that a human is expected to trust.
## Output
```
### Wiki Sweep
- Diff scope: {N files, M hunks | no source files — skipped}
- wiki/architecture.md: {NO-IMPACT | PATCH — reason, citing file:line from the diff}
- wiki/data-models.md: {…}
- wiki/critical-paths.md: {…}
- wiki/external-services.md: {…}
- rules/{each file}: {NO-IMPACT | PATCH — which bullet is wrong, citing file:line}
### Proposed Removals
{one line per section the diff appears to retire — file, what, why — or `none`}
```
Append BOTH blocks — `### Wiki Sweep` and `### Proposed Removals` — to the
change's `espalier/changes/{type}/{slug}/doc-patches.md`, creating it if it is
not there. That file is the sweep's record: `delivery_brief` reads the
`### Wiki Sweep` block out of it for the brief's `## Doc Refresh` section, and
a removal that exists only in your reply is a removal nobody will ever action.
A sweep whose verdicts live only in the transcript leaves the human approving
at Stage 10 with no idea a doc was touched.
Keep each verdict line exactly in that shape — `- wiki/<file>.md: PATCH — …`
or `- rules/<file>.md: PATCH — …`, path first, the word `PATCH` right after
the colon. It is not only for the human: `espalier-pr.sh project-sweep` reads
the LATEST `### Wiki Sweep` block and commits exactly the files marked
`PATCH` there, nothing else. A file you edit without a `PATCH` line is never
committed; a `PATCH` line for a file you did not edit commits nothing. On a
re-run (a rollback back through Stage 7), append a fresh block — never edit
the earlier one.
Then apply the patches you proposed — edit the files directly — and stop.
Committing them is Stage 7's job, not yours: the orchestrator calls
`espalier-pr.sh project-sweep wiki` and `project-sweep rules`, so each class
lands as its own revertable commit on the espalier PR. That call refuses a doc
that was deleted or lost more lines than it gained, so a removal applied here
stops Stage 7 instead of shipping.
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: 198k | 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 |