Gate tool calls per command and path, not per tool name

Approval was a list of tool names: `bash` needed it, `read_file` did not. That
fails in a specific way. `bash` covers `git status` and `rm -rf` equally, so a
user working through a batch presses `a` — always allow — on the first prompt and
every later command runs unasked, including the one they would have refused. The
gate was strongest when it mattered least and gone by the time it mattered.

Rules now match the *subject* of a call: the command for `bash`, the path for a
file tool, the pattern for a search.

    "permission": {
      "bash": { "*": "ask", "git *": "allow", "rm *": "deny" },
      "edit_file": { "*": "deny", "src/generated/*": "allow" }
    }

Plain last-match-wins, with no special case for deny. An earlier version made
deny win wherever it sat, on the theory that a refusal should be impossible to
undo by accident, and it made default-deny-with-exceptions unexpressible — which
is the shape a careful user actually writes, and the same shape as `*.env` denied
while `*.env.example` is allowed. Refusals that must never be configurable stay
in the guard plugin, which runs ahead of this and which --yolo cannot reach.

Three behaviours fall out of it:

- `always` grants the pattern the tool suggests, not the tool. Approving
  `git status` runs `git log` unprompted and still asks about `npm publish`.
- `.env`, `.env.*`, and `.pem` are denied on read by default. Not gated, refused:
  a secret that reaches the context is on the wire and in the session file, and
  there is no taking it back. `.env.example` stays allowed.
- A call repeated identically three times in one turn asks even when allowed. A
  model repeating itself is not making progress, and `bash: allow` is a statement
  about which commands are safe rather than permission to loop.

The prompt now says which rule matched and what `always` would grant:

    bash wants to run
    git status --porcelain
    y allow once | a always allow bash git * | n deny

A typo in a decision string is dropped at parse time, leaving the tool on its
default. Treating an unparseable value as `allow` would mean one misspelling
silently removing the gate.

Written after surveying Claude Code, Codex, opencode, and phi. Three of the four
had already moved to per-pattern rules; the credential deny and the repeat guard
come from opencode directly. ROADMAP records what was deliberately not taken and
why — OS sandboxing needs three platform implementations and is worse than
nothing if half-built, and opencode's own docs say LSP integration is often not a
net positive.

581 tests, up from 572. The engine is a pure function tested on its own, and the
loop is tested through a real Session: an allowed pattern never prompts, a denied
one never executes, and a `.env` read leaves no secret in the transcript.

README.md

@@ -51,7 +51,7 @@ agent: default  thinking: medium
 cwd: /home/you/project
 skills: debug, refactor, review, test
 plugins: guard, time
-approvals: on for write_file, edit_file, multi_edit, bash, mcp__*
+approvals: ask for write_file, edit_file, multi_edit, bash, mcp__*
 /help for commands
 
 > why does the pagination test fail?
@@ -64,11 +64,12 @@ installed and honours `.gitignore`. `list_dir` gives an ignore-aware tree so it
 blindly to orient, and `read_many_files` pulls a batch in one round trip. `read_file` refuses
 binaries rather than filling the context with mojibake.
 
-**Edits with your approval.** Every `write_file`, `edit_file`, `multi_edit`, and `bash` call
-stops for a `y`/`a`/`n` decision, with a coloured diff for edits. `multi_edit` is atomic, so
-a failing match leaves the file untouched rather than half-changed. The `guard` plugin
-refuses irreversible commands outright — `rm -rf`, `git reset --hard`, force pushes,
-`DROP TABLE` — and `--yolo` cannot bypass it.
+**Edits with your approval, gated per command.** `write_file`, `edit_file`, `multi_edit`, and
+`bash` stop for a `y`/`a`/`n` decision, with a coloured diff for edits. Rules match the command or
+path rather than the tool, so `git *` can run unprompted while everything else still asks —
+answering `a` whitelists that pattern, not the whole tool. `.env` and `.pem` files are refused on
+read outright. The `guard` plugin refuses irreversible commands ahead of any of it — `rm -rf`,
+`git reset --hard`, force pushes, `DROP TABLE` — and `--yolo` cannot bypass it.
 
 **Shows its work.** Reasoning streams to a collapsed panel you can expand with `ctrl-r`, the
 tool in flight is named as it runs, and `bash` output streams live instead of arriving all at
@@ -113,6 +114,7 @@ the flags are.
 |---|---|
 | [Configuration](docs/configuration.md) | config file, provider presets, environment, every flag |
 | [Tools](docs/tools.md) | every tool, tool sets and what they cost, the approval model |
+| [Permissions](docs/permissions.md) | allow/ask/deny rules, patterns, defaults, the repeat guard |
 | [Agents and thinking](docs/agents.md) | variants, thinking levels, step caps, which to reach for |
 | [Skills](docs/skills.md) | the bundled skills, writing your own, why the catalogue is split |
 | [Plugins](docs/plugins.md) | the interface, the guard and its limits, builtin versus installed |