@event4u/agent-config 1.15.0 → 1.17.0

package/.agent-src/rules/language-and-tone.md

@@ -17,155 +17,72 @@ NO MOMENTUM EXCEPTION. NO TECHNICAL-CONTEXT EXCEPTION.
 NO "SWITCH MID-PARAGRAPH". NO "LAST 20 TURNS WERE ENGLISH".
 ```
 
-**Overrides** conversation momentum, tool-output habits, prior-reply
-language, codebase language, open-file language, files-just-edited
-language, convenience. First thing to check on every reply, last thing
-to check before sending.
-
-Canonical failure: agent edited English `.md` for many turns; user
-types short German (`3`, `weiter`, `mach das`, `und jetzt X`); agent
-answers English because momentum wins. **Trigger is the user's last
-message, not the turn count.** Length irrelevant — `3` after a German
-question still means German continues.
-
-### Source of language truth — chat messages ONLY
-
-```
-THE LANGUAGE SIGNAL IS THE USER'S CHAT MESSAGES. NOTHING ELSE.
-OPEN FILES, ROADMAPS, .md CONTENT, TOOL OUTPUT, CODE, COMMIT MESSAGES,
-TICKETS, PR DESCRIPTIONS, FILE NAMES — NONE OF THEM COUNT.
-```
-
-`.md` files in this repo are English by rule (see below) — that says
-nothing about chat language. Same for: file contents read via `view` /
-`grep`, quoted commits / tickets / PRs / branches, code identifiers,
-the agent's own previous replies. Only the most recent **chat message**
-sets the language. User opens an English roadmap and types German →
-reply in German.
+Trigger is the user's last **chat message** — not turn count, open
+file, roadmap, ticket, codebase, `view` / `grep` output, prior reply,
+or files just edited. Short German (`3`, `weiter`, `mach das`) after
+many English turns flips the reply to German.
 
 ### Pre-send gate — MANDATORY before every reply
 
-Run silently **before** emitting any tokens:
-
-1. **Detect** — language of user's last **chat message** (not the open
-   file, not the roadmap, not the prior reply).
-   German signals: "ich", "Du", "nicht", "warum", "wie", "ist", umlauts. <!-- md-language-check: ignore -->
-   English signals: "I", "you", "is", "the", "how".
-   Mixed → mirror the **dominant** language; tie → German wins (project default).
-2. **Check** — is drafted prose (not code, not file contents) in that language?
-3. **Rewrite** — if no, rewrite whole prose before sending. No exceptions, no
-   "just this sentence", no "the technical term is English anyway".
-4. **Confirm** — first sentence must be in target language. No English opener
-   before switching mid-paragraph.
-
-### The rule, spelled out
-
-- User writes German → **MUST** respond in German (informal "Du", never "Sie").
-  "Du" capitalized at sentence start, lowercase otherwise.
-- User writes English → respond in English.
-- User switches mid-conversation → switch on the **very next** reply. No
-  grace period, no "let me finish this thought in the old language".
-- Code blocks, command output, file contents, quoted tool output stay in
-  their native language. Only the **prose around them** mirrors the user.
-- "I've been answering in English for a while" is NOT a reason to keep going.
-  Trigger is the **latest user message**, not conversation momentum.
-- Numbered option lists (per `user-interaction`) mirror the user's language —
-  `.md` source is English, rendered reply is translated at runtime.
-
-### When the user calls out a language slip
-
-1. Acknowledge **once**, briefly, in the correct language ("Entschuldigung" /
-   "Sorry"). One sentence, no excuses.
-2. Switch immediately on the same reply.
-3. Do **not** re-explain the mistake in the wrong language.
-4. Do **not** promise "from now on" — just do it. Only behaviour changes
-   prove compliance.
-5. If user asks to harden the rule, harden it on this turn — don't defer.
-
-### Failure modes to watch for
-
-- Drafting reply in English first, then "translating the intro" → English
-  phrasing with German words. Draft in target language from the first token.
-- Copy-pasting English option labels from `.md` sources without translating.
-- Mixing languages inside a table or bullet list because "the technical term
-  is English" — surrounding prose must still mirror. Keep proper nouns and
-  code identifiers as-is; translate everything else.
-- Assuming English because "the codebase is English" — codebase language ≠
-  conversation language.
-- Mirroring the **open file** the IDE reports — open files are background
-  context, not chat messages.
-- Mirroring the **roadmap or ticket** being executed — artefacts are English
-  by `.md` rule; chat language is whatever the user wrote.
-
-## Other language rules
-
-- All code comments must be in English.
-- All `.md` documentation files must be in English (see section below). If
-  an existing file is in German, translate it when you touch it.
-- Use two spaces after icons like ❌, ✅, ⚠️ in CLI output. One space is not enough. For other icons, one space is fine.
-- Avoid double and triple blank lines in code and output — one blank line is enough.
-- Every file MUST end with exactly one newline — no trailing blank lines.
+Run silently before any output:
 
-## `.md` files are ALWAYS English — no exceptions
+1. **Detect** — language of the user's last chat message. Mixed →
+   **dominant** language; tie → German (project default).
+2. **Check** — is drafted prose (not code, not file contents) in
+   that language?
+3. **Rewrite** — if no, rewrite whole prose. No "just this sentence",
+   no "technical term is English anyway".
+4. **Confirm** — first sentence in target language. No English opener
+   before switching mid-paragraph.
 
-**Every** piece of text inside `.md` files in `.augment/` and `agents/` must be in English.
-This includes:
+### Spelled out
 
-- Headings, paragraphs, and bullet points
-- **Example option labels** (e.g., `> 1. Yes — start implementing`, NOT `> 1. Ja — mit der Umsetzung starten`)
-- **Example prompts and questions** (e.g., `"Found X unresolved comments."`, NOT `"X offene Kommentare gefunden."`)
-- **Template placeholders and sample output** (e.g., `Progress:`, NOT `Fortschritt:`)
-- **ASCII art labels** in formatted output blocks (e.g., `CHANGES:`, NOT `ÄNDERUNGEN:`)
-- **Table headers and content**
+- German → respond in German, informal "Du" (never "Sie"). "Du"
+  capitalized at sentence start, lowercase otherwise.
+- Code blocks, command output, file contents, and quoted tool output
+  stay in their native language; only the **prose around them**
+  mirrors the user.
+- Numbered options (per `user-interaction`) — `.md` source English,
+  rendered reply translated at runtime.
 
-The agent translates to the user's language **at runtime** when presenting options.
-The `.md` source files are the English blueprint — they define WHAT to say, not in which language.
+### Slip handling
 
-**Wrong** (German in `.md`):
-```
-> 1. Interaktiv — bei jedem Kommentar nachfragen
-> 2. Automatisch — alle selbstständig abarbeiten
-```
+Acknowledge **once**, briefly, in the correct language
+("Entschuldigung" / "Sorry"). Switch on the same reply. Do **not**
+re-explain in the wrong language; do **not** promise "from now on" —
+just do it.
 
-**Correct** (English in `.md`):
-```
-> 1. Interactive — ask before each comment
-> 2. Automatic — handle all independently
-```
+Full failure-mode list and wrong-vs-correct examples in
+[`../../docs/guidelines/agent-infra/language-and-tone-examples.md`](../../docs/guidelines/agent-infra/language-and-tone-examples.md).
 
-### Quoted user-input examples — same rule, with one labeled exception
+## Other language rules
 
-Common drift pattern: a rule documents trigger phrases the agent
-should recognize and writes them as quoted German examples inside
-English prose. **Not allowed**, even when demonstrative.
+- Code comments in English.
+- `.md` documentation files in English (see section below).
+  Translate existing German `.md` files when you touch them.
+- Two spaces after icons `❌`, `✅`, `⚠️` in CLI output; one space
+  for other icons.
+- One blank line max; no double or triple blank lines.
+- Every file ends with exactly one newline.
 
-**Wrong** (DE quote embedded in EN prose):
+## `.md` files are ALWAYS English — no exceptions
 
-```md
-Single-decision delegation ("für diesen Schritt entscheide du") →
-handle that step autonomously.
+**Every** piece of text inside `.md` files in `.augment/`,
+`.agent-src/`, `.agent-src.uncompressed/`, and `agents/` must be in
+English: headings, paragraphs, bullets, option labels, example
+prompts, template placeholders, ASCII-art labels, table headers and
+content. The agent translates to the user's language **at runtime**.
 
-A standing "arbeite selbstständig" never lifts the floor.
-```
+### Labeled-anchor exception
 
-**Correct** — two ways:
+Quoting German examples inside English prose is **not allowed**.
+Two correct ways:
 
-1. **Translate to English.** Trigger recognition is semantic; the
-   agent matches intent across languages regardless of the example.
-   `("you decide for this step")` works as well as the German.
+1. **Translate to English** — trigger recognition is semantic.
 2. **Use a labeled `DE: … · EN: …` anchor list** when the
-   multilingual nature of recognition is the point:
-
-   ```md
-   - DE: "arbeite selbstständig" · "frag nicht jedes Mal" · "tue es einfach"
-   - EN: "work autonomously" · "don't ask" · "just do it"
-   ```
-
-The labeled-anchor block is the **only** allowed location for German
-prose in an English `.md` file. Body text, example sentences, prompt
-templates, agent-rendered strings, and failure modes must be English.
-Reference established phrases abstractly later (e.g. "a standing
-autonomy directive") and link back to the anchor block.
+   multilingual nature of recognition is the point. This is the
+   **only** allowed location for German prose in an English `.md`;
+   reference phrases abstractly elsewhere and link to the anchor.
 
 ### Detection heuristic
 
@@ -180,4 +97,4 @@ Before saving an `.md` file under `.augment/`, `.agent-src/`,
 - Non-English quoted phrases in body text (paragraphs, list items,
   table cells) when the surrounding prose is English.
 
-Hit → translate the fragment or move it into a `DE: … · EN: …` block.
+Hit → translate it or move it into a `DE: … · EN: …` block.

package/.agent-src/rules/minimal-safe-diff.md

@@ -1,7 +1,7 @@
 ---
-type: "always"
-alwaysApply: true
-description: "Minimal safe diff — the smallest change that solves the stated problem; no drive-by edits, no opportunistic refactors, no reformatting of untouched code"
+type: "auto"
+alwaysApply: false
+description: "When writing or reviewing a diff — the smallest change that solves the stated problem; no drive-by edits, no opportunistic refactors, no reformatting of untouched code"
 source: package
 ---
 

package/.agent-src/rules/missing-tool-handling.md

@@ -6,34 +6,38 @@ description: "When a CLI tool needed for the task is not installed — ask befor
 
 # Missing Tool Handling
 
-When a CLI tool is needed and **not installed** (`command not found`,
-`which X` empty), **STOP and ask** before working around it or installing it.
+When a CLI tool is needed to solve the task cleanly and is **not installed**
+(`command not found`, `which X` empty, missing from `$PATH`), **STOP and ask**
+before working around it or installing it.
 
 ## The rule
 
-- **Never install silently.** Installing changes the user's system —
-  requires explicit permission (`scope-control`).
+- **Never install silently.** Installing is an action that changes the user's
+  system — it requires explicit permission (see `scope-control`).
 - **Never silently work around** a missing tool with a brittle substitute
-  (awk for YAML, `grep` for JSON, string splicing) when a proper tool is
-  the standard answer. The workaround hides the dependency.
-- **Ask with numbered options** (`user-interaction`). State the tool, why
-  it fits, the install command, and the workaround cost.
+  (awk-regex for YAML, `grep` for JSON, string splicing) when a proper tool
+  is the standard answer for the job. The workaround hides the real
+  dependency and makes the decision invisible to the user.
+- **Ask with numbered options** (see `user-interaction`). State which tool,
+  why it's the best fit, the install command, and the workaround cost.
 
 ## When it applies
 
-- Shell returns `command not found` for a tool the task genuinely needs
-  (yq, jq, rtk, gh, docker, mkcert, terraform, …).
-- A skill or spike needs the idiomatic tool but it's absent locally.
-- You're about to substitute a verbose script for a single tool call
+- Any shell command fails with `command not found` for a tool the task
+  genuinely needs (yq, jq, rtk, gh, docker, mkcert, terraform, …).
+- A skill or spike requires a tool that is the idiomatic answer but is
+  absent locally.
+- You are about to substitute a verbose script for a single tool invocation
   because the tool isn't there.
 
 ## When it does NOT apply
 
-- Nice-to-have — clean substitute already present (e.g. `jq` available →
-  no `yq` needed just for JSON).
-- Tool is forbidden by project policy (check `scope-control` and
-  tool-allowlists first).
-- Missing library dependency → use `composer`, `npm`, `pip` per package
+- The tool is a nice-to-have and a clean substitute already exists in the
+  repo (e.g. `jq` present → no need for `yq` just for JSON).
+- The tool is explicitly forbidden by project policy (check
+  `scope-control` and tool-allowlists first).
+- The missing artefact is a library dependency — those go through
+  `composer require`, `npm install`, `pip install` per the package-manager
   rules, still with explicit permission.
 
 ## How to ask
@@ -49,14 +53,16 @@ When a CLI tool is needed and **not installed** (`command not found`,
 > 4. Skip this path — I propose a different approach
 ```
 
-After the user picks: if **install**, wait for confirmation or run the
-documented command only if explicitly authorised this turn. If
-**workaround**, record the decision in the artefact (comment or ADR).
+After the user picks: if **install**, wait for confirmation that it is
+done (or run the documented install command only if the user explicitly
+authorises it in this turn). If **workaround**, record the decision in
+the artefact (comment or ADR) so the cost is visible.
 
 ## Capture the learning
 
-If the same tool keeps missing across tasks, flag it in the project's
-setup docs or `.agent-settings.yml` prerequisites.
+If the same tool keeps coming up as missing across multiple tasks, flag
+it in the project's setup docs or `.agent-settings.yml` prerequisites.
+Don't make every session re-discover the gap.
 
 See also: `scope-control` · `ask-when-uncertain` · `user-interaction` ·
 `tool-safety`.

package/.agent-src/rules/model-recommendation.md

@@ -1,7 +1,7 @@
 ---
-type: "always"
-description: "Model recommendation — detect task complexity, suggest optimal model (Opus/Sonnet/GPT), check on task changes and command invocations"
-alwaysApply: true
+type: "auto"
+description: "Starting a new task, switching task type, or invoking a command — detect task complexity and recommend the optimal model (Opus/Sonnet/GPT) before any work"
+alwaysApply: false
 source: package
 ---
 
@@ -30,7 +30,7 @@ If the detected model is `gemini`, immediately trigger the Gemini warning (see b
 `/commit`, `/fix-pr-comments` are sonnet tasks — even if the previous task was opus-level.
 This is the most commonly missed check. Do NOT skip it.
 
-**Priority over commands rule**: This check runs BEFORE the `slash-commands` rule. If a model switch
+**Priority over commands rule**: This check runs BEFORE the `slash-command-routing-policy` rule. If a model switch
 is recommended, show the suggestion first. Only after the user responds, execute the command.
 
 **If ambiguous** (could be opus or sonnet): default to **sonnet** — cheaper, and the user can escalate.

package/.agent-src/rules/no-cheap-questions.md

@@ -0,0 +1,82 @@
+---
+type: "always"
+description: "No cheap questions — never ask what context answers, never offer Iron-Law-violating options, never stage no-trade-off choices; mode-independent (off / auto / on)"
+alwaysApply: true
+source: package
+---
+
+# No Cheap Questions
+
+A question is **cheap** when the answer follows from stated context,
+an option breaches an Iron Law, choices differ only in sequencing /
+format, or one option is obviously dominant. Cheap questions are
+noise, regardless of `personal.autonomy`.
+
+Mode-independent. [`autonomous-execution`](autonomous-execution.md)'s
+"trivial" failure modes scope to `personal.autonomy: on` (or
+`auto`-after-opt-in); this rule lifts the **no-trade-off** subset to
+`off` and pre-opt-in `auto` too.
+
+## The Iron Laws
+
+```
+NEVER ASK WHAT THE STATED CONTEXT ALREADY ANSWERS.
+NEVER PRESENT AN OPTION THAT VIOLATES AN IRON LAW.
+NEVER OFFER NUMBERED CHOICES WITHOUT A REAL TRADE-OFF.
+```
+
+Hold in `off`, `auto`, and `on`. Autonomy never lifts them.
+
+## What counts as cheap
+
+- **Sequencing** — "Step 2 or 3 next?" when the roadmap orders them.
+- **Format-only** — "Table or paragraph?"; no semantic trade-off.
+- **Commit asks** — forbidden by [`commit-policy`](commit-policy.md).
+- **CI / test asks** — [`verify-before-complete`](verify-before-complete.md) decides, not the user.
+- **Fenced-step re-asks** — "Start Phase 1?" after *"plan only"*; see
+  [`scope-control § fenced step`](scope-control.md#fenced-step--user-set-review-gates).
+- **Iron-Law option** — option breaches `commit-policy`,
+  `scope-control § git-ops`, or `non-destructive-by-default`.
+- **Context-derived** — answer follows from prior turn / standing
+  instruction / roadmap; act, state the assumption inline.
+- **Dominant option** — one choice obviously correct; alternatives
+  carry no upside.
+- **Re-ask after decline** — forbidden per
+  [`scope-control § decline = silence`](scope-control.md#decline--silence--no-re-asking-on-the-same-task).
+
+Concrete examples per class:
+[`asking-and-brevity-examples`](../../docs/guidelines/agent-infra/asking-and-brevity-examples.md#cheap-question-class-catalog--extended-examples).
+
+## Pre-Send Self-Check — MANDATORY before every question
+
+Run silently before any numbered-options block:
+
+1. Answer already in stated context?
+2. Any option violates [`commit-policy`](commit-policy.md), [`scope-control § git-ops`](scope-control.md), or [`non-destructive-by-default`](non-destructive-by-default.md)?
+3. Options pure sequencing / format, no trade-off?
+4. One option obviously dominant?
+5. User fenced next step (*"plan only"*, *"review first"*) → deliver + handback per [`scope-control § fenced step`](scope-control.md#fenced-step--user-set-review-gates).
+6. User already declined? Re-ask forbidden per [`scope-control § decline = silence`](scope-control.md#decline--silence--no-re-asking-on-the-same-task).
+
+Any "yes" → **do not ask**. Pick the dominant path, state assumption
+inline (*"assuming X — adjust if wrong"*), hand back. One-question-per-turn
+from [`ask-when-uncertain`](ask-when-uncertain.md) still applies when
+the question is genuine.
+
+## When asking IS allowed
+
+- Real architectural / scope decision with non-obvious trade-offs.
+- Vague-request trigger per [`ask-when-uncertain`](ask-when-uncertain.md#vague-request-triggers--must-ask).
+- Security-sensitive path per [`security-sensitive-stop`](security-sensitive-stop.md).
+- Hard Floor in [`non-destructive-by-default`](non-destructive-by-default.md) — confirmation mandatory.
+- Two genuinely-equivalent paths; user preference is the tiebreaker.
+
+In doubt → ask. This rule narrows asking, never widens silence.
+
+## Interactions
+
+- [`ask-when-uncertain`](ask-when-uncertain.md) — vague triggers + one-question-per-turn; narrows the cheap subset.
+- [`autonomous-execution`](autonomous-execution.md) — mode-scoped triviality there; mode-independent floor here.
+- [`commit-policy`](commit-policy.md) · [`scope-control`](scope-control.md) · [`non-destructive-by-default`](non-destructive-by-default.md) — Iron Laws this rule defends.
+- [`user-interaction`](user-interaction.md) — numbered-options shape; this rule decides whether to send.
+- [`direct-answers`](direct-answers.md) — brevity, no flattery.

package/.agent-src/rules/no-roadmap-references.md

@@ -0,0 +1,73 @@
+---
+type: "auto"
+description: "Adding a link to a specific file in agents/roadmaps/ from any stable artifact (rule, skill, command, context, guideline) — roadmaps are transient; promote durable findings to agents/contexts/ instead"
+alwaysApply: false
+source: package
+---
+
+# No Roadmap References from Stable Artifacts
+
+Roadmaps in `agents/roadmaps/` are **transient** — archived, skipped,
+or deleted as work completes. Stable artifacts (rules, skills,
+commands, contexts, guidelines, AGENTS.md, README) outlive them. A
+stable artifact citing a specific roadmap file becomes a broken
+reference the moment that roadmap is deleted.
+
+## The Iron Law
+
+```
+NEVER LINK TO A SPECIFIC FILE IN agents/roadmaps/ FROM A STABLE ARTIFACT.
+PROMOTE DURABLE CONCLUSIONS TO agents/contexts/ AND CITE THAT INSTEAD.
+```
+
+Stable artifact = anything that is **not** a roadmap, council
+session, chat-history archive, commit message, or PR description.
+
+## Forbidden patterns
+
+These paths must not appear inside a stable artifact:
+
+- `agents/roadmaps/<file>.md`, `agents/roadmaps/archive/<file>.md`,
+  `agents/roadmaps/skipped/<file>.md`
+
+Stable artifact = any file under `.agent-src.uncompressed/{rules,
+skills,commands,contexts,templates,personas}/`, `agents/contexts/`,
+`docs/guidelines/`, `docs/contracts/`, `docs/architecture.md`,
+`docs/customization.md`, `docs/getting-started.md`, `docs/catalog.md`,
+`AGENTS.md`, `README.md`, `copilot-instructions.md`.
+
+CI enforcement: `scripts/check_no_roadmap_refs.py` (companion linter
+— fails the build on any new violation).
+
+## Allowed patterns
+
+- `agents/roadmaps/` and its subdirectories as directory mentions
+  (talking about the layer, not a specific file)
+- Roadmap → roadmap references (siblings within the transient layer)
+- Council sessions, `.agent-chat-history`, commit messages, PR
+  descriptions — transient by construction, not part of the package
+  surface
+
+## What to do instead
+
+When a stable artifact needs to cite a roadmap finding:
+
+1. Identify the durable conclusion — decision, contract, lesson,
+   mechanic.
+2. Promote it to a context file under `agents/contexts/` (ADR,
+   mechanics doc, locked decision). The roadmap can then point at
+   the context, not the other way around.
+3. Reference the context from the stable artifact.
+
+Failure mode: *"I'll just link to the roadmap, it's evidence."* The
+roadmap gets archived, then deleted, then the link rots. **Promote
+first, link second.**
+
+## See also
+
+- [`docs-sync`](docs-sync.md) — cross-reference sync after rename / delete
+- [`agent-docs`](agent-docs.md) — roadmap layer conventions
+- [`roadmap-progress-sync`](roadmap-progress-sync.md) — sync dashboard on
+  roadmap touch
+- [`augment-source-of-truth`](augment-source-of-truth.md) — edit
+  `.agent-src.uncompressed/`

package/.agent-src/rules/non-destructive-by-default.md

@@ -1,8 +1,10 @@
 ---
 type: "always"
-description: "Agent is never destructive and never endangers production systems — Hard Floor that always asks for production-trunk merges, deploys, pushes, prod data/infra changes, whimsical bulk deletions, and commits containing bulk deletions or infra changes; no autonomy setting, roadmap step, or standing instruction can bypass it"
+description: "Agent is never destructive — Hard Floor always asks for prod-trunk merges, deploys, pushes, prod data/infra, bulk deletions, and bulk-deletion/infra commits; no autonomy or roadmap bypass"
 alwaysApply: true
 source: package
+load_context:
+  - .agent-src.uncompressed/contexts/authority/destructive-mechanics.md
 ---
 
 # Non-Destructive by Default
@@ -51,57 +53,21 @@ of truth makes it reversible. Lives in
 ## Bulk deletions during WIP — allowed if task-connected
 
 Deletions inside an **active, user-stated task** are allowed in the
-working tree, **even multiple files or multiple folders**. The floor
-moves to the **commit** (row 6 above), not the in-progress edit.
-
-**Allowed during WIP (no floor on the edit):**
-
-- A roadmap step or current task explicitly names the files / folders to remove
-- Refactor naturally drops deprecated code the user already agreed is dead
-- Cleanup of generated artifacts — `node_modules/`, `dist/`, `.next/`,
-  build caches, `vendor/` reinstall — never source code
-- Single-file edits, single-class refactors, deleting **one** file the
-  user just named
-- Renames and moves (technically delete + add)
-
-**Floor fires on the edit when the deletion is:**
-
-- Whimsical — "while I was in there", drive-by cleanup not part of the task
-- Unnamed scope — "delete all the old tests" without a list, glob
-  across unrelated files, "clean up the legacy folder" with no inventory
-- ≥5 unrelated files in one operation, outside the current task scope
-- Content destruction — `DROP TABLE`, `TRUNCATE`, `git reset --hard`
-  past unpushed work, database wipes (destroys *content*, not just tree)
-
-Ambiguous → floor wins. Ask.
-
-**The commit of task-aligned bulk deletions still needs its own ask.**
-A roadmap or task authorizes the *edit*; only the user-this-turn
-authorizes the *commit* (row 6). Surface the diff (paths + counts), get
-confirmation, then commit.
+working tree, **even multiple files or multiple folders** — the Hard
+Floor moves to the **commit** (row 6 above), not the in-progress edit.
+Whimsical / drive-by / unnamed-scope deletions still trip the floor on
+the edit. Full allowed/forbidden lists in
+[`destructive-mechanics`](../contexts/authority/destructive-mechanics.md)
+§ Bulk deletions during WIP.
 
 ## Failure modes
 
-- Treating a standing autonomy directive as cover for a Hard-Floor
-  action. Standing autonomy never lifts the floor; merging to `main`,
-  deploying, pushing, prod-data edits, or whimsical `rm -rf <dir>`
-  always ask.
-- Reading a roadmap step that says "deploy to staging" or
-  "merge into main" as authorization. The roadmap can sequence the
-  work; only the user-this-turn can authorize the floor crossing.
-- Refusing to delete files the user already named because "the floor
-  fires on `rm`". It does not — task-aligned WIP deletions are
-  allowed, even multi-folder. The floor fires when the deletion is
-  whimsical, unscoped, or about to be committed.
-- Committing a diff that removes a directory, deletes ≥5 unrelated
-  files, or touches Terraform / k8s manifests / Ansible without
-  surfacing the diff first — even when [`commit-policy`](commit-policy.md)
-  otherwise authorizes commits (e.g. `/commit-in-chunks`, roadmap
-  pre-scan, an explicit "commit this now"). Bulk-deletion / infra
-  commits need their own ask, every time.
-- Reading a roadmap step listing files to delete as authorization to
-  *commit* the deletion. The step authorizes the *edit*; the commit
-  is row 6 of the Hard Floor and needs its own confirmation.
+The full failure-mode catalog (autonomy-as-cover, roadmap-as-authorization,
+refusing-named-deletions, commit-without-diff-surface,
+roadmap-step-≠-commit-authorization) lives in
+[`destructive-mechanics`](../contexts/authority/destructive-mechanics.md)
+§ Failure modes. Reach for it when a Hard-Floor situation feels
+ambiguous; the rule itself stays focused on the trigger table.
 
 ## Cloud Behavior
 

package/.agent-src/rules/onboarding-gate.md

@@ -1,7 +1,7 @@
 ---
-type: "always"
-description: "First-run gate — when onboarding.onboarded is false in .agent-settings.yml, prompt the user to run /onboard before anything else"
-alwaysApply: true
+type: "auto"
+description: "First turn of a conversation on a project — check onboarding.onboarded in .agent-settings.yml; when false, prompt the user to run /onboard before executing any other request"
+alwaysApply: false
 source: package
 ---
 
@@ -73,7 +73,7 @@ gate. This protects projects that were set up before this rule shipped.
   conversation, max.
 - Replace normal settings edits. Mid-life changes are ad-hoc (edit the
   file directly or ask the agent, which follows
-  [`layered-settings`](../guidelines/agent-infra/layered-settings.md#section-aware-merge-rules));
+  [`layered-settings`](../../docs/guidelines/agent-infra/layered-settings.md#section-aware-merge-rules));
   this rule is a one-time gate.
 - Run on every agent turn. First turn only.
 
@@ -89,6 +89,6 @@ gate. This protects projects that were set up before this rule shipped.
 ## See also
 
 - [`/onboard`](../commands/onboard.md) — the command this gate invokes
-- [`layered-settings`](../guidelines/agent-infra/layered-settings.md) — merge rules for mid-life edits
+- [`layered-settings`](../../docs/guidelines/agent-infra/layered-settings.md) — merge rules for mid-life edits
 - [`agent-settings` template](../templates/agent-settings.md) — `onboarding.onboarded` reference
 - [`rule-type-governance`](rule-type-governance.md) — why this is `always`