---
name: peaks-sop
description: 'Authoring skill for user-defined SOPs (standard operating procedures) in Peaks. Use when a user wants to create, edit, debug, or register their own gated workflow — ordered phases plus gates that block advancement until conditions are met — by describing it in natural language instead of hand-writing JSON or memorizing CLI commands. DOMAIN-AGNOSTIC: not just software/release flows — equally for content publishing, compliance and approval checklists, data pipelines, onboarding, ops runbooks, or any personal repeatable procedure, wherever "don''t enter the next stage until X is true" applies and X is checkable via a file, file content, or a command.'
---

<!-- peaks:loop-hygiene — HAND-MAINTAINED block, NOT generated; byte-identical in every SKILL.md, enforced by tests/unit/skills/loop-hygiene-block.test.ts. Edit one, edit all 23. -->

## Loop hygiene (every turn — MANDATORY)

**Skill header.** While this skill is active, open every turn with
`Peaks-Loop Skill: <skill> | Peaks-Loop Role: <role> | Peaks-Loop Gate: <gate> | Peaks-Loop Context: <ratio%> (<zone>) | Next: <one short action>`.
Every turn, not only the first — it is how the user knows which skill is driving.
`Role` and `Context` are machine facts: copy them from `peaks skill presence --json`
(`data.role`, `data.context.ratioPct`, `data.context.action`) — never invent them.

**Context is this skill's own business.** Run `peaks skill presence --json` every turn and
read its `context` block. When `action` is `auto-fire`, `pre-compact`, or `red-line`,
run `peaks code auto-compact --project .` **yourself**, then continue. Tell the user the
ratio in one line if it helps, but never hand them the compaction step: asking the user to
run `/compact` is the regression the zero-pause contract forbids. This holds in **every
mode** — standard *and* 24h. The mode moves the threshold, never the obligation.

**Read before you edit.** Read a file before your first `Edit` / `Write` / `MultiEdit` on
it — the normal way to work here — for every path outside `.peaks/**` (source, tests, docs,
config); `.peaks/**` writes are exempt.

**Expect one denial per file anyway — and do NOT pre-empt it.** A `PreToolUse` gate (ECC's
"Fact-Forcing Gate") denies the FIRST edit a session makes to any given path, once, by
design. Reading does NOT prevent it: the gate keys on the path's first touch, not on whether
you read it. **Do not recite its four questions before every edit** — it asks when it wants
them, and reciting unprompted burns a round-trip per file for nothing. Answer only when a
denial actually arrives.

**When one does arrive:** a denial is not a failure and the tool is not broken — your edit
was NOT applied. State the facts it asks for (importers, affected API, data schemas if any,
the user's verbatim instruction), then retry the SAME operation. The retry is allowed. Do not
switch tools, do not give up, do not retry blindly. One more thing worth knowing: an idle gap
of ~30 minutes clears the gate's "already passed" list, so a file you cleared earlier can be
denied again after a long pause. That is the gate resetting, not you regressing.
<!-- /peaks:loop-hygiene -->

# Peaks-Loop SOP Authoring

Peaks-Loop SOP turns a natural-language workflow description into a validated, registered custom SOP, then helps the user debug it until each gate behaves as intended. The user describes the process in plain language; this skill drives the `peaks sop` CLI on their behalf — they never have to hand-write `sop.json` or remember the command sequence.

**This is a general workflow-gating tool, not a developer-only tool.** A SOP is any repeatable process with ordered stages where you must not skip ahead until conditions are met. Software release is just one example; content publishing, compliance/approval checklists, data validation pipelines, employee onboarding, ops runbooks, and personal procedures are all first-class — often the more valuable use. When you interview the user, do not assume code: ask about *their* process in *their* domain.

## What to tell the user BEFORE running any command

This skill helps you create, test, and enforce a repeatable workflow (a "SOP")
with gates that physically block advancement until conditions are met. It works
for ANY domain — content publishing, compliance, onboarding, data pipelines,
software releases, personal procedures.

Before you run `peaks skill presence:set` or any other setup command, tell the
user (in your own words, one or two sentences):

  "I'll help you turn your process into a Peaks SOP — ordered stages plus gates
   that make sure you never skip a step. I'll interview you about your workflow,
   generate the definition, test it, and register it. Ready?"

If the user has never used Peaks before, offer a 30-second demo:

  "Want me to create a quick demo SOP first? It takes 30 seconds — I'll
   scaffold one, show you how a gate blocks, then we can replace it with your
   real workflow."

When they accept, use `demo-sop` as the id (NOT `peaks-sop` — the `peaks-` prefix
is reserved), name it "Demo SOP", make two phases (draft → done), one gate
(file-exists README.md), then run the full init→lint→check→register→hooks
install flow so they see the whole thing. Then offer to replace it with their
real process.

## Skill presence (MANDATORY first action — run AFTER telling the user what you're doing)

If the user invoked this skill directly ("peaks-sop") without mentioning a
specific project, assume `--project .` (current directory). If they are NOT in
the middle of a multi-skill peaks-code/prd/rd/qa pipeline, skip `statusline
install` — the status bar is for long-running orchestrations; a standalone SOP
authoring session does not need it.

```bash
peaks skill presence:set peaks-sop --project <repo> --mode <mode> --gate startup
peaks project memories --project <repo> --json
```

Then display: `Peaks-Loop Skill: peaks-sop | Peaks-Loop Role: <role> | Peaks-Loop Gate: startup | Peaks-Loop Context: <ratio%> (<zone>) | Next: <one short action>`. `Role` and `Context` are machine facts: copy them from `peaks skill presence --json` (`data.role`, `data.context.ratioPct`, `data.context.action`) — never invent them. Update gates with `peaks skill presence:set peaks-sop --project <repo> --mode <mode> --gate <gate>` when they change. When the SOP is registered and the user is satisfied, run `peaks skill presence:clear --project <repo>`.

## Responsibilities

- interview the user to turn a natural-language workflow into ordered **phases** and the **gates** that guard entry into each phase;
- generate a valid SOP manifest (`sop.json`) and the registrable `SKILL.md` via the CLI — never make the user hand-author JSON;
- run the lint → fix → re-lint debug loop until the manifest is clean;
- test each gate (pass/fail/blocked) and dry-run advancement so the user sees the SOP behave before committing;
- register the SOP so it joins skill presence / statusline like a built-in skill;
- explain the three gate types and the security posture of command gates.

## What a SOP is (explain this to the user)

A SOP is an **ordered list of phases** plus **gates** bound to phases. A gate that does not pass blocks advancement into its phase — this is how "don't drop steps" applies to the user's own workflow. The SOP lives at `.peaks/sops/<sop-id>/` (manifest `sop.json` + a registrable `SKILL.md`). Gate checks come in three types:

| type | fields | passes when |
|------|--------|-------------|
| `file-exists` | `path` | the file exists (path pinned inside the project) |
| `grep` | `file` + `pattern` (+ `absent`) | the regex matches in the file — or, with `absent: true`, does NOT match |
| `command` | `run` (argv array) + `expectExitZero` | the command exits as expected |

Prefer **`grep` with `absent: true`** for any "must not contain X" gate (no leftover `TODO`, no placeholder, no unresolved marker). It is a pure-text check — no `--allow-commands`, cross-platform, no shell. Reach for a `command` gate only when the check genuinely needs to run a program.

`command` gates run user-defined processes and are **refused by default** — they require explicit `--allow-commands`, run with no shell (argv array, no injection), a timeout, and cwd pinned to the project. Always tell the user when a SOP needs `--allow-commands` and why.

### Where SOP files live (two definition layers, per-project execution)

A SOP *definition* (manifest + SKILL.md) can live in two layers:
- **Global** `~/.peaks/sops/<sop-id>/` — your personal SOPs, reusable across every project. `init` / `lint` / `register` default here (no `--project`).
- **Project** `<project>/.peaks/sops/<sop-id>/` — **committed into the repo**, so a teammate who clones it gets the SOP (and, with the hook installed, is enforced). Pass `--project <repo>` to `init` / `lint` / `register` to use this layer.

The **project layer takes precedence** over global for the same id. Execution reads see the merged view (project wins). A SOP's *run-state* (current phase, history) is always **per-project**: `<project>/.peaks/sop-state/<sop-id>/`. `check` / `advance` take `--project` (default: current directory) to say which project the gates evaluate against, whose progress advances, and which definition layer wins.

`advance` also enforces **phase order**: you may re-enter the current phase, step back, or move to the immediately-next phase, but not skip ahead — a forward jump returns `SOP_PHASE_SKIP` (bypassable, like a gate, with `--allow-incomplete --reason`).

### Where SOPs apply (lead with the user's domain, not code)

The three gate primitives are domain-neutral, so the same engine governs very different workflows:

| domain | phases (example) | gate idea |
|--------|------------------|-----------|
| content / publishing | draft → edit → publish | `file-exists` the draft; `grep` no `TODO`/`TKTK`; `command` runs a spell/word-count check |
| compliance / approval | prepare → review → sign-off | `file-exists` `approval.md`; `grep` the doc contains "Approved" |
| data pipeline | raw → cleaned → validated | `command` runs a validator script that exits 0 |
| onboarding / ops | request → provision → done | `file-exists` each checklist artifact; `command` verifies a config |
| personal procedure | any repeatable steps | whatever "don't forget step X" means, expressed as a file/grep/command |

The one boundary to explain: a gate must reduce to **a file existing, text matching in a file, or a command's exit code**. A purely human-judgment gate ("did the editor approve?") is expressed by reifying it into a signal — e.g. require an `approved.md` file, or that a status file contains "approved". The `command` gate is the universal adapter for anything scriptable.

### Literal-word trap and `stripMeta` (added by PRD 006 on 2026-06-02)

A naive `grep` gate for a content-publishing SOP has a self-referential failure mode: the author writing *about* the gate's pattern ("we use a `grep absent: "TODO"` gate to block leftover T-O-D-O") ends up triggering the gate they just described. This is rare in code-review SOPs and very common in content/publishing SOPs.

**Opt-in workaround**: add `stripMeta: true` to the gate's check. The evaluator strips HTML comments (`<!-- … -->`), fenced code blocks (` ``` … ``` `), and C-style block comments (`/* … */`) from the file content *before* applying the regex. The author's discussion of the gate in those areas is no longer counted.

```json
{
  "id": "no-todo",
  "phase": "publish",
  "check": { "type": "grep", "file": "post.md", "pattern": "TODO", "absent": true, "stripMeta": true }
}
```

Default `false` preserves byte-identical behavior for existing SOPs. The stripper is conservative on malformed input (unclosed fences / block comments fall through un-stripped). Not covered by this slice: inline code (`` `TODO` ``) and blockquotes (`> TODO`) — both are content the author explicitly chose to render, so they remain subject to the regex. If a future dogfood surfaces a need to strip those too, open a follow-up PRD.

## Un-bypassable enforcement (optional, opt-in)

By default a SOP gate only blocks the `peaks sop advance` command — nothing forces the agent through it. To make a gate **physically un-bypassable**, a SOP can declare **guards** that bind a concrete irreversible Bash action to a phase, and the user installs a PreToolUse hook:

```jsonc
// in sop.json
"guards": [ { "phase": "publish", "bash": "git +push" } ]
```
Meaning: running a Bash command matching `git +push` IS entering the publish phase, so publish's gates must pass first. Then:

```bash
peaks hooks install --project <repo>     # explicit, opt-in; writes one PreToolUse entry
```

Now when the agent tries `git push` while the publish gate fails, Claude Code receives `permissionDecision: "deny"` and the command is blocked **before any permission check — it holds even under `--dangerously-skip-permissions`**. CI only blocks at merge; CLAUDE.md instructions are advisory; this blocks in-conversation and cannot be skipped.

- `bash` is a **JS regex inside JSON** — escape backslashes (`"git\\s+push"`) or just use `"git +push"`. `peaks sop lint` rejects an invalid regex (`GUARD_INVALID_PATTERN`).
- Emergency override: `peaks gate bypass --sop <id> --phase <phase> --reason "<why>"` records a **one-shot** token consumed by the next blocked command (capped per project per SOP, reason audited).
- Trust: enforcement **fails open** — any internal error allows the command (a Peaks bug never bricks Claude Code); only a real gate failure denies. Installing the hook is an explicit user command; this skill never writes `settings.json` itself.
- `peaks hooks status` / `peaks hooks uninstall` manage the hook.

> Team enforcement: register the SOP into the **project layer** (`peaks sop init/register --project <repo>`) so the definition is committed in the repo. A teammate who clones it — even with an empty global `~/.peaks` — is enforced by the same gates once they install the hook. (A SOP that lives only in your global `~/.peaks` enforces only on your machine.)

## Default runbook

The default sequence this skill executes on the user's behalf. The natural-language → generate → debug loop IS this runbook.

```bash
# self-check (required by doctor; no user-visible effect)
peaks skill runbook peaks-sop --json

# 1. interview the user FIRST (before any scaffold): what are the ordered stages?
#    For each stage, what must be true before entering it? Translate answers into
#    file-exists / grep / grep-absent / command gates.

# 2. scaffold the SOP based on the interview (preview first, then apply)
#    definitions are global (~/.peaks/sops) — init/lint/register take no --project
peaks sop init --id <sop-id> --name "<human name>" --json
peaks sop init --id <sop-id> --name "<human name>" --apply --json

# 3. write the phases/gates from the interview into ~/.peaks/sops/<sop-id>/sop.json
#    (edit the manifest directly — the user described it in natural language)

# 4. DEBUG LOOP: lint, fix the reported findings, re-lint until clean
peaks sop lint --id <sop-id> --json
peaks sop lint --id <sop-id> --allow-commands --json   # when the SOP uses command gates

# 5. test each gate behaves as intended (pass / fail / blocked) against a project
peaks sop check --id <sop-id> --gate <gate-id> --project <repo> --json

# 6. dry-run the flow to confirm gates + phase order block/allow the right phases
peaks sop advance --id <sop-id> --to <phase> --project <repo> --dry-run --json

# 7. register the SOP (preview, then apply) so it joins presence/statusline
peaks sop register --id <sop-id> --dry-run --json
peaks sop register --id <sop-id> --json
peaks sop registry --json

# 8. (optional) make a gate un-bypassable: declare guards in sop.json, then install the hook
peaks hooks install --project <repo>
peaks hooks status --project <repo>
#    emergency one-shot override when a guarded action must proceed despite a failing gate:
peaks gate bypass --sop <sop-id> --phase <phase> --reason "<why>" --project <repo>

# 9. hand the SOP to the user; clear presence when done
peaks project memories:extract --session-id <session-id> --project <repo> --json  # extract durable memories
peaks skill presence:clear --project <repo>
```

### Transition verification gates (MANDATORY — run the command, see the output)

You cannot declare a SOP ready from memory. Each gate below is a command you **MUST run** and whose output you **MUST see**.

**Peaks-Loop Gate A — the manifest lints clean before register:**
```bash
peaks sop lint --id <sop-id> --json
# Expected: ok:true. If ok:false, fix the findings and re-lint — do NOT register a SOP that does not lint.
```

**Peaks-Loop Gate B — gates behave as intended before handing off:**
```bash
peaks sop check --id <sop-id> --gate <gate-id> --project <repo> --json
# Confirm each gate returns the verdict the user expects (pass on the good state, fail/blocked otherwise).
```

## Debugging guidance

When `sop lint` reports findings, fix them in `sop.json` and re-lint — common findings: duplicate gate id, gate bound to an undefined phase, missing check fields, or a command gate without `--allow-commands`. When `sop check` returns `blocked`, the check could not be evaluated (path escaped the project, target file unreadable, command not permitted or failed to spawn) — distinct from `fail` (evaluated, condition not met). Use `sop advance --dry-run` to confirm the blocking behaves before any real advance.

Concrete manifest reference, gate cookbook, and the debug loop: `references/sop-authoring.md`.

## Boundaries

Do not implement the user's business code, run their real release, or modify runtime configuration. This skill authors and validates the SOP definition; the SOP's gates then govern the user's own workflow. `command` gates execute user-defined commands only with explicit `--allow-commands` consent. Do not register a SOP that does not lint clean.
