@r1z4x/tezgah 1.9.2 → 1.13.0

package/docs/skills.md

@@ -11,11 +11,30 @@ install step that links them. The always-on rule text a session is told is
 A skill is `skills/<name>/SKILL.md`: YAML frontmatter carrying `name` and
 `description`, then a markdown body. The body is instructions for the model - the
 read tool opens it, nothing executes it. `plan-add` and `ponytail` also declare
-`argument-hint` in that frontmatter (`skills/plan-add/SKILL.md:11`,
-`skills/ponytail/SKILL.md:16`), and Claude exposes each skill as
+`argument-hint` in that frontmatter (`skills/plan-add/SKILL.md:7`,
+`skills/ponytail/SKILL.md:6`), and Claude exposes each skill as
 `/tezgah:<name>` - a bare `/plan-add` is a command that does not exist
 (`tests/test_skills.py::SkillStandards.test_slash_commands_carry_the_prefix_the_plugin_registers`).
 
+Every host shows the `description`, and hosts cut it.
+The omp host keeps about twelve words.
+Codex and Claude drop descriptions once the skill list passes their budget. So a shipped description opens with `Use when`, names its
+English and Turkish triggers, and stays within 160 characters
+(`tests/test_skills.py::DESCRIPTION_CAP`).
+The routing word `tests/routing-fixtures.md` pins for the skill sits in its first
+100 characters (`tests/test_skills.py::DESCRIPTION_KEPT`).
+The longer text a description used to carry opens the body, under `## When to use`.
+A skill body that calls a tezgah CLI writes its installed path,
+`~/.config/tezgah/bin/<cmd>`: no host but Claude puts that directory on PATH.
+
+A vendored library entry nested under a skill is `ENTRY.md`, not `SKILL.md`.
+Codex and Cursor list every `SKILL.md` below a skills directory, nested ones
+included. There are 144 vendored entries: 98 in `ai-research`, 39 in
+`design-library`, 5 in `rl-env` and 2 in `pm-frameworks`.
+Those hosts listed each one as a skill, which pushed codex past its skills budget. Each keeps its bytes and upstream path under the new
+name, and `bin/tezgah-import-ai-research` writes `ENTRY.md` on a re-vendor
+(`bin/tezgah-import-ai-research::vendored`).
+
 Two neighbours are not skills. A **slash command** is a file under `commands/`
 wired only through the Claude plugin channel - `commands/ponytail.md` and
 `commands/adhd.md` are shells over `tezgah-pony` and `tezgah-adhd`, and no other
@@ -67,8 +86,11 @@ A line is cut at 140 characters with a trailing `...`; a description with no
 trigger sentence keeps its first sentence.
 
 The on-demand file also carries a **collision table**: the pairs among tezgah's
-own skills whose descriptions share three or more subject words, computed from
-the router lines themselves (`skill_collisions()`, `bin/tezgah-setup::skill_collisions`).
+own skills whose `## When to use` trigger lines share three or more subject words
+(`skill_collisions()`, `bin/tezgah-setup::skill_collisions`).
+`skill_when_line()` reads that line from the body (`bin/tezgah-setup::skill_when_line`).
+The short bilingual description shares too few words for two near-neighbours to pair.
+The longer `## When to use` text still tells them apart.
 It rides `opencode-skills.full.md` and never the always-on file, because it is
 read when a session is about to choose, not every turn. It is a view, not a
 rename: it names `plan-add` / `plan-status` / `plan-sync` and
@@ -124,9 +146,37 @@ CLI into `~/.config/tezgah/bin` beside the other tezgah commands.
 On a turn the judgement does not answer, the same search runs on its own,
 default on and with no model (`hooks/tezgah_skill_pick.py::section_hint`). It
 keeps its index as SQLite in the tezgah cache dir (`skill-search.sqlite`). It
-rebuilds that index only when a SKILL.md's mtime or size moves
-(`hooks/tezgah_skill_pick.py::_open_index`). A prompt reads its own terms'
-postings and the length column, never the corpus.
+rebuilds that index only when a SKILL.md's mtime or size moves, or when an older
+`INDEX_VERSION` wrote the index
+(`hooks/tezgah_skill_pick.py::_open_index`, `hooks/tezgah_skill_pick.py::INDEX_VERSION`).
+A prompt reads its own terms' postings and the length column, never the corpus.
+
+The hint reads the project's skills from the session's cwd (`section_hint(..., cwd=)`).
+`project_roots()` finds the nearest `.git` top and takes its `.claude/skills`,
+`.agents/skills` and `skills/` directories (`hooks/tezgah_skill_pick.py::project_roots`).
+A project that ships its own skills gets its own index file,
+`skill-search-<hash>.sqlite`.
+Two sessions in two such projects then do not rebuild each other's index on every turn.
+
+Ranking runs in two tiers (`hooks/tezgah_skill_pick.py::search`).
+tezgah's own skills and the project's rank first.
+The hosts' machine-wide directories (omp's, Claude's, `~/.agents`, Codex's,
+opencode's) answer only when the first tier cannot (`hooks/tezgah_skill_pick.py::_second_tier`).
+Before the split, 16 of 54 live hints in one transcript set pointed at cyber skills.
+In another set, 3 of 7 pointed at marketing skills. Both sets came from product repositories.
+The search covers `tezgah-contract`, but the hint never names it (`hooks/tezgah_skill_pick.py::HINT_SKIP`).
+Replayed over 1331 owner prompts, it still took 76 of 123 hints.
+
+A skill's frontmatter description rides its first section twice
+(`hooks/tezgah_skill_pick.py::DESC_WEIGHT`).
+On the 18 English owner prompts of the adoption set, top-1 moved from 10 to 16.
+Every term folds Turkish letters to ASCII on both sides (`ç ğ ö ş ü` to `c g o s u`).
+A word longer than five letters also indexes its five-letter prefix
+(`hooks/tezgah_skill_pick.py::STEM`, `hooks/tezgah_skill_pick.py::terms`).
+A Turkish word carries case and possessive in suffixes, so "tasarımını" meets
+"tasarım" only through the prefix. Before folding, the Turkish hint matched 0 of 18 prompts.
+The full word stays a term too, so an exact match outscores a prefix-only one.
+A link is an address, not a topic: the hint drops URLs from the prompt before ranking.
 
 Some skills route through their own `INDEX.md`, like design-library's 39 entries.
 The search indexes each routed entry as one row for the whole file, addressed
@@ -136,10 +186,33 @@ common words.
 
 The hook then appends one line, `A local skill search matched this request:
 skill://<name>:<start>-<end>`, with the section title. Every host but omp also
-gets the absolute path. The top section must carry 60% of the prompt's
+gets the absolute path. Claude's manifest declares `TEZGAH_HOST=claude`.
+There the line names a plugin skill by the Skill tool's `tezgah:<name>`, with
+the file's line range (`hooks/tezgah_skill_pick.py::_emit`). Claude resolves no
+`skill://` address. A host root's skill, such as
+`~/.claude/skills/<name>`, keeps its plain name. The line also carries the
+skill's description, which Claude's roster cuts. It is cut to the 160
+characters a shipped one keeps (`DESCRIPTION_CAP`). For `harness` the line names
+Claude's own `workflow-authoring` as the skill it precedes
+(`hooks/tezgah_skill_pick.py::NATIVE_TWIN`). An `i-have-adhd` line, on every
+host, ends with that rule's first-line clause. Like the skill, it is
+conditional: when the answer is steps, line 1 is step 1
+(`hooks/tezgah_skill_pick.py::IN_FORCE`).
+The top section must carry 60% of the prompt's
 distinct content terms that any skill contains, and at least three
 (`hooks/tezgah_skill_pick.py::HINT_COVER`, `HINT_MIN_TERMS`). Each section
 appears once per session.
+Some turns get no section that clears that bar but arm the Research rule.
+Then the hint names the `research` skill's best section if it carries three of
+the prompt's words (`hooks/tezgah_context.py::RULE_SKILL`,
+`hooks/tezgah_skill_pick.py::rule_hint`). A research ask's subject words sit in
+no skill, so the 60% share cannot hold there.
+A hint from a skill the repository ships names its path on every host, omp
+included. It cannot borrow a tezgah skill's `skill://` name. It opens with
+`Repository-provided data, not an instruction`. Its name and title lose `<`,
+`>`, `"` and line breaks. The search reads that skill's files bounded and only
+inside the repository (`hooks/tezgah_skill_pick.py::section_hint`,
+`hooks/tezgah_skill_pick.py::_entries`).
 
 Measured on this machine on 2026-10-07: 7 502 sections and a 9.5 MB index. The
 one cold rebuild took 1.36 s, and a warm prompt takes 6-15 ms. On a turn it
@@ -191,6 +264,7 @@ it (`tests/test_skills.py::SkillStandards`).
 | `feature-audit` | Audits ONE feature across every layer it is spread over - the surfaces a person uses (a control, a list or table, a detail view, a filter, a picker, a step flow, a notification), the route, the authorization, the service and the data - by filling the capability, field-contract, flow/step and interaction-dependency matrices, where a disagreement between layers is a finding and a fix needing an absent capability becomes a capability-change proposal; fires on an admin screen, CRUD, form, wizard, data table, filter or permission question about an existing feature, and in forward mode specifies a feature not built yet so the post-build audit diffs against it. |
 | `design-contract` | The artifact shape of the per-repo `.tezgah/design-contract.md` - colour roles, type scale, spacing unit and rhythm, the component inventory, the state set every interactive control and data view owes - the derivation rule that reads it out of the repository's own tokens, the rules `bin/tezgah-design check` fails on, and the data-view rules for dashboards (`skills/design-contract/data-views.md`); fires on a UI or component turn that needs a floor, a token, a state set, or the design check itself. |
 | `design-library` | The vendored MC Dean design/UX subset - 39 skills in the `cognitive-accessibility`, `adaptive-interfaces`, `inclusive-interaction`, `visual-critique` and `design-research` plugins - read one entry at a time through its own `INDEX.md`, never the tree; fires when a UI or product turn needs a cognitive-load, inclusive-interaction or visual-critique method tezgah has no counterpart for. |
+| `hallmark` | The vendored UI generator - one ask, then a whole page: 16 Floor rules, 48 themes, 20 macrostructures and its own `sloplint` gate list, with the material locked into a root `design.md`; fires when the turn builds a new page or theme rather than judging an existing one. |
 | `ai-research` | The vendored 98-skill library for AI/ML machinery - training and serving a model, benchmarks, interpretability, retrieval pipelines; read one entry, never the tree. |
 | `rl-env` | The vendored FineEnvs skill set for authoring an RL environment from a description - OpenEnv, OpenReward/ORS, Verifiers and NeMo Gym, plus the orchestrator that ports one description across all four; read one entry, never the tree. |
 
@@ -236,7 +310,7 @@ adaptation changed - the same record an added adaptation is expected to carry.
 `skills/ai-research` is the large vendored tree: 98 upstream skills in 23
 categories, bodies byte-for-byte upstream including frontmatter and `author`
 lines, at revision `773a52944ba4747a18bd4ae9ade53fff041adcbc`
-([skills/ai-research/SOURCE:1-20]). No machine-scraped reference dumps and no
+([skills/ai-research/SOURCE:1-22]). No machine-scraped reference dumps and no
 LaTeX template trees were copied, and every drop is named in that file.
 `bin/tezgah-import-ai-research --check` verifies the tree against its manifest.
 
@@ -270,17 +344,58 @@ trap cases the library is scored against. Two bodies depart from upstream, and
 `skills/design-library/SOURCE` and `NOTICE` name both: the touch-target entry
 paired `44x44 CSS pixels` with `WCAG 2.2 Level AA`, where SC 2.5.8 is Level AA at
 24x24 and SC 2.5.5 is Level AAA at 44x44 - 24 is the floor `bin/tezgah-design`
-enforces - and `adaptive-personalisation/SKILL.md` carried the sibling
+enforces - and `adaptive-personalisation/ENTRY.md` carried the sibling
 `contextual-help-design` skill's `name`, which a host that loads a skill from its
 directory skips. `tests/test_skills.py` hashes every listed body, fails on an
 unlisted directory, and fails on a body that pairs 44x44 with Level AA again.
 
+`skills/hallmark` is the generator, vendored whole: 178 upstream files at
+revision `7ec8682` (2026-10-22), MIT, with upstream's `LICENSE` beside its
+`SOURCE` ([skills/hallmark/SOURCE:1-19]). `SOURCE` carries a sha256 for every
+file and the checkout command for a re-vendor. `tests/test_skills.py::HallmarkVendor`
+fails on a drifted body, a missing listed file, and a file the table leaves out.
+Its `SKILL.md` is upstream's workflow and ships as upstream's bytes. `SOURCE`
+lists it too, because tezgah writes no entry point of its own for this tree.
+That listing is what exempts the file from the frontmatter lint
+(`tests/test_skills.py::Frontmatter`). The name sits in `SKILLS`
+(`bin/tezgah-setup::SKILLS`), so every host links it and the router lists it.
+
+## The design layer: two floors
+
+A UI turn answers to two floors. Hallmark's `design.md` owns the material:
+genre, shape, theme, the OKLCH palette, the font roles and the motion character.
+The contract owns the state set, the component inventory, the measured evidence
+and the facts. Neither floor may overrule the other by silence.
+
+`derive` reads a root `design.md` or `DESIGN.md` before any token file, writes
+its palette as hex and stamps `source: design-md`
+(`bin/tezgah-design::design_md_tokens`). The contract's `precedence` block is
+the only tie-break, and the code reads it rather than a session. `ours` keeps
+states, components, evidence and facts. `theirs` hands material and motion
+character to Hallmark. `shared` covers contrast, focus-visible, reduced motion
+and the tap target, where both sides want the same thing and the numbers stay
+ours.
+
+`precedence_problem` refuses an ambiguous table with exit 2. It refuses a key
+named twice, `facts` handed to a waiver or to the user's words, and a `verbatim`
+value other than `style` (`bin/tezgah-design::precedence_problem`).
+`tezgah-design slop` is the verb that merges both verdicts. It runs Hallmark's
+vendored gate list and then the contract check over the same tree. It writes
+`.tezgah/design/slop.json` and stays not clean when either floor fails, when the
+render tier never ran, or when no measurement judged a component
+(`bin/tezgah-design::slop_cli`). The Stop rule reads that file as UI evidence
+beside `measurement.json` (`hooks/tezgah_integrity.py::DESIGN_SLOP`). The long
+form, the state set and the rest of the contract shape live in the skill
+(`skills/design-contract/SKILL.md:103-130`).
+
 ## Adding a skill
 
 1. Write `skills/<name>/SKILL.md` with frontmatter carrying `name` and a
-   `description` whose **later** sentence begins `Use when ...` - that sentence
-   is the router line, and it must carry the words a session would match on
-   (`bin/tezgah-setup::skill_description`).
+   `description` that opens `Use when ...`, names its English and Turkish
+   triggers and stays within 160 characters. That sentence is the router line,
+   and it must carry the words a session would match on
+   (`bin/tezgah-setup::skill_description`). Put the longer text in the body's
+   first section, `## When to use`. The collision table reads it.
 2. Add the name to `SKILLS` (`bin/tezgah-setup::SKILLS`). `SKILLS` drives the
    router, the linking, the uninstall and the budget; a name without a
    `SKILL.md`, or a directory without a `SKILLS` entry, is not a shipped skill.
@@ -318,6 +433,15 @@ unlisted directory, and fails on a body that pairs 44x44 with Level AA again.
   its index and its three trap cases
 - `skills/rl-env/SOURCE` - the vendored FineEnvs environment-authoring set, its
   upstream revision and the per-file sha256
+- `skills/hallmark/SOURCE` - the vendored generator, its revision and a sha256
+  per file
+- `bin/tezgah-design` and its `.py` twin - `derive`, `design_md_tokens`,
+  `precedence_problem`, `check` and `slop_cli`: the two floors and the one verb
+  that merges them
+- `skills/design-contract/SKILL.md` - the contract shape, the state set and the
+  precedence classes
+- `hooks/tezgah_integrity.py` - `DESIGN_SLOP`, the Stop rule's reader of
+  `.tezgah/design/slop.json`
 - `bin/tezgah-skill` - the section search CLI over every installed skill root;
   `hooks/tezgah_skill_pick.py` - `sections`, `search`, `skill_roots`,
   `section_of`, the hint that carries the address

package/docs/status-line.md

@@ -25,9 +25,13 @@ hosts/dsh/statusline/lib/client.js:86-90).
 | 1 | `consult`, `research`, `graph`, `orch`, `judge` | the on-demand capabilities |
 | 2 | `idx` | a per-repo fact: code-graph readiness |
 | 3 | `plans` | a per-repo fact, its own group |
+| ≥4 | a repository's [extension](#a-repositorys-own-marks-extension) chips | one cluster per extension, after every built-in |
 
 Groups 0 and 1 come from `_GROUP` (`hooks/tezgah_context.py::_GROUP`); the prefix
-carries -1 and `idx` and `plans` carry 2 and 3 where they are appended (`hooks/tezgah_context.py::version_segment`, `hooks/tezgah_context.py::health_segments`). Outside every
+carries -1 and `idx` and `plans` carry 2 and 3 where they are appended (`hooks/tezgah_context.py::version_segment`, `hooks/tezgah_context.py::health_segments`). An
+extension declares no group. `health_segments()` puts its chips in the group after
+the highest built-in one, so a chip never splits the armed/used run
+(`hooks/tezgah_context.py::health_segments`). Outside every
 [root](glossary.md#root) the checklist still prints — tezgah loads globally on
 opencode, so the indicator must not go silent — and only the per-repo extras are
 omitted (`hooks/tezgah_context.py::health_segments`, `tests/test_statusline.py::test_outside_roots_still_shows_segment`).
@@ -124,6 +128,94 @@ instead of forking (`hooks/tezgah_context.py::health_segments`). `plans` is `pla
 plan files carry `status: blocked` in their first 400 bytes — `ready` with a
 blocked plan, else `info` (`plan_mark()`, `hooks/tezgah_context.py::plan_mark`).
 
+### A repository's own marks (extension)
+
+A repository adds chips to the line without tezgah naming it. It drops one
+manifest at `<repo>/.tezgah/extension.json`. That manifest names a command
+printing structured segments on stdout:
+
+```json
+{"statusline": {"command": ["bin/my-statusline", "--segments"], "timeout_ms": 300}}
+```
+
+`segments()` walks up from the working directory to the first manifest
+(`manifest_path()`, `hooks/tezgah_extension.py::manifest_path`). It runs the
+command once per render, with the repo root as cwd. It appends what comes back
+after every built-in mark (`segments()`, `hooks/tezgah_extension.py::segments`,
+`hooks/tezgah_context.py::health_segments`). A relative `command[0]` resolves
+against the repo root, so a manifest travels with its clone.
+
+The walk stops after `MAX_WALK` levels. Past that, a working directory under
+`$TEZGAH_ENGAGEMENTS_ROOT/<slug>` still reads `<slug>/.tezgah/extension.json`
+(`_engagement_manifest()`, `hooks/tezgah_extension.py::_engagement_manifest`).
+An unset root, a cwd outside it and an engagement without a manifest cost one
+`isfile` and no chips. `tezgah-status --manifest=FILE` names the manifest
+outright (`given_manifest()`, `hooks/tezgah_extension.py::given_manifest`). A
+FILE that is not there costs the chips, never the line.
+
+Every call hands the child its own envelope on stdin, as one JSON object:
+
+```json
+{"api": 1, "event": "statusline", "cwd": "/repo/sub", "repo_root": "/repo",
+ "session_id": "s-42"}
+```
+
+An extension therefore reads which repository and which session asked it, rather
+than inferring that from its own environment (`call_envelope()`,
+`hooks/tezgah_extension.py::call_envelope`). `api` is the envelope's version: an
+extension that meets a higher one says so on the line rather than drawing a chip
+it cannot trust. A command that never reads its stdin sees the same closed, empty
+input it always did.
+
+The contract is **data, not a line**. The extension produces `{key, state, glyph,
+text}` segments, and tezgah draws them. So color, the narrow-surface drop and the
+group separators stay identical for every extension. An extension never has to
+know the gutter. It must name `text` and `state`. tezgah fills a missing `glyph`
+from the state's own mark, so a chip reads like a built-in. The per-extension cap
+(`MAX_SEGMENTS`) and the hard timeout (`timeout_ms`, default
+`DEFAULT_TIMEOUT_MS`) bound one row.
+
+It is a subprocess and not an import on purpose. The extension is the user's
+repo, in whatever language, and may be slow or broken. A bounded child keeps a
+broken extension costing one chip on one line, never the turn. Every way it can
+fail answers no segments. A missing manifest, an absent command, a non-zero exit,
+a timeout and malformed output all cost the extension its chips
+(`run_manifest()`, `hooks/tezgah_extension.py::run_manifest`, `tests/test_extension.py`). A repo with no
+manifest pays one `os.path.isfile` per level and nothing else.
+
+tezgah-cyber is the first extension. Its `bin/engagement-statusline --segments`
+turns an engagement's `notes/status.json` into `cyber-eng`/`cyber-needs`/
+`cyber-running`/`cyber-steps`/`cyber-verified`/`cyber-severity` chips. An operator
+then reads an engagement's state at a glance, in the line they already read.
+
+#### The manifest's other keys
+
+The same manifest carries four more keys. One function in
+`hooks/tezgah_extension.py` reads each, and one caller consumes it. Each reader
+checks its value the way `statusline` does. A wrong type, a malformed entry or
+an unreadable manifest costs that key or that entry, never the line or the gate.
+
+```json
+{"checks": ["./bin/run-tool", "bin/audit --strict"],
+ "untrusted": ["nmap", "sqlmap"],
+ "agents": [{"match": "(?i)\\brecon\\b", "agent": "recon"}],
+ "statusline": {"command": ["bin/my-statusline"], "override": true}}
+```
+
+| Key | Read by | Consumed by | Effect | Fail posture | Bound |
+|---|---|---|---|---|---|
+| `checks` | `repo_checks()` | `hooks/tezgah_integrity.py::repo_checks`, folded into `_check_runs` | a shell command whose words start with one prefix is a check, so its run records `verify_ok`/`verify_fail` and `passing_check` accepts a pass | fail-open to the built-in vocabulary: a non-list, a non-string entry or one shlex rejects is dropped; a prefix that is one of the five no-op programs `echo`, `true`, `cat`, `printf`, `ls` alone (any path to them, `/bin/true` too) is dropped, since its run would license a pass that checked nothing, while further words after the program (`echo tested`) keep it a check (`hooks/tezgah_extension.py::NO_OP_PROGRAMS`); a dropped prefix costs its credit, never the gate; a prefix inside a quoted argument or a quoted heredoc body (`<<'EOF'`) is no run, while an unquoted heredoc's body (`<<EOF`) is read as commands, so a prefix line there counts as a run - the built-in vocabulary's own ceiling (`hooks/tezgah_integrity.py::_repo_check_runs`); a check whose status a pipe or a later `;` command owns records as ran, as `pytest; echo` does | `MAX_CHECKS` entries |
+| `untrusted` | `untrusted_programs()` | `hooks/tezgah_integrity.py::untrusted_source` | a shell command that runs a listed program (at a command position, `shell_programs`) is the `repo` channel: its result is labelled and the next effect carries the taint notice, as a `network` read does | fail-open to no channel: a non-list or a non-name entry is dropped | `MAX_UNTRUSTED` names |
+| `agents` | `repo_agent()` | `hooks/tezgah_models.py::route` (`_repo_route`) | after the harness judgement, the first entry whose `match` regex finds the brief names the agent; the tier, `via` and the judge stay the harness's, and a brief the override rule pinned keeps its agent | fail-open to the harness pick: a pattern `re` rejects, a non-string pattern or an agent that is no one-word name is dropped | `MAX_AGENTS` entries, `MAX_PATTERN` characters a pattern, `MAX_BRIEF` characters of brief read |
+| `statusline.override` | `override()` | `bin/tezgah-status` | literally `true`: the project owns its line and points its own `statusLine` at `tezgah-status`, which then reads the host payload on stdin (`cwd`/`workspace.current_dir`, `session_id`) when no PATH is given; the built-in marks stay and the chips follow them | anything but `true` is no override and stdin is not read; a malformed payload draws the line from the process directory | `PAYLOAD_MAX` bytes, `PAYLOAD_WAIT` seconds on a silent pipe |
+
+The `checks` and `untrusted` readers need the call's directory. The PostToolUse
+row passes its `cwd`, and the gate's `began` row passes its own, so `_pair`
+finds the pass. Lacking one, the untrusted reader uses the call's `cwd`
+argument, else the hook's own directory. opencode's in-process
+classifier (`hosts/opencode/plugins/tezgah.js`) does not read the manifest.
+There a declared program stays unlabelled: a missed label, never a false one.
+
 ### The two skill-read marks, and `--observable=`
 
 `pony` and `adhd` mean "the full skill text reached the session" — the only signal
@@ -207,6 +299,8 @@ Every other surface reads the store.
   segment from the active theme and rebuilding the separator from `seg.group`
   (56-88,hosts/opencode/tui/tezgah-tui.tsx:108-111). Its `tezgah: status legend` command opens a dialog running
   `--legend` (141-147).
+  On opencode 2 the same line comes from hosts/opencode/tui/cli/tui.tsx, a
+  `Plugin.define` CLI plugin listed in `cli.json`. It reuses the v1 file's `run`.
 - **dsh** — hosts/dsh/statusline/lib/index.js serves `GET /api/tezgah.status`
   behind the Web UI's own authenticated fetch fence, so the browser fetches it
   same-origin with no extra token handling (14,34-38); it always passes
@@ -217,6 +311,12 @@ Every other surface reads the store.
   legend on hover, pins it on click (16-19,35,47-52,63-64).
 - **Codex** — hosts/codex/hook.py puts the plain line in `systemMessage`
   (182-184,195-198).
+- **Amp** — the plugin draws the widest plain tier
+  (`hosts/amp/hook.py::status_line`) on Amp's experimental status item
+  (`createStatusItem`, TUI only, plain text). It draws only for the thread the
+  user is looking at. It redraws on `session.start`, `agent.end` and each
+  watched tool result, and shows `hook✗` while the hook fails. A build without
+  the API has no status surface (`hosts/amp/tezgah-plugin.ts.in`).
 
 ## Seeing and debugging the line
 
@@ -227,6 +327,7 @@ see a line by hand:
 tezgah-status /path/to/repo <session-id>     # the line for one repo and session
 tezgah-status /path/to/repo <id> --json      # segments: key/state/glyph/text/group
 tezgah-status /path/to/repo <id> --counters  # the evidence ledger's counters
+tezgah-status --rates --since A --until B    # the ledger folded between two dates
 tezgah-status --legend                       # what each mark means
 tezgah-status --rule-yield                   # live gate rules that stopped firing
 tezgah-status --enforcers                    # what enforces each always-on rule
@@ -259,6 +360,25 @@ three places: no session id (used marks stay `○`), a surface that passed
 `--observable=` and so renders dim where you expected a state, or a store a
 sandboxed host could not write (`hooks/tezgah_paths.py::cache_dir`).
 
+`--rates --since DATE --until DATE` is the windowed reader over the same
+ledgers (`rates`, `hooks/tezgah_integrity.py::rates`). It folds the rows stamped
+in `[since, until)` per window and per calendar day. The fold carries turns,
+`claim` rows and the Stop classes that blocked them. It also carries the shape
+flags of each judged reply, and denies by rule. The rates are blocks per 100
+turns and denies per 100 sessions. They also carry the claim block and Stop
+shape shares, and blocked claims per 100 sessions by class. `counters_all` still
+totals a lifetime on purpose, and the window is this reader's, so the two cannot
+disagree about a row. The dates are local calendar days. `until` is exclusive,
+which makes two adjacent windows a partition of the span rather than an
+approximation of one. A repeated `--since`/`--until` pair prints one block per
+window. It also prints the pp delta between adjacent windows, the unit the
+observation rows in `.tezgah/plans` state their bars in. A window that read more
+than one row contract prints the split. A delta whose two windows lean on
+different contracts says so. There, `claims` is every judged reply under `v2`
+and only a reply that claimed under `v3` (plan 135). `--json` prints
+`{"windows": [...]}` - one shape for one window and for many. A window that
+cannot be read, or `--since`/`--until` without `--rates`, exits 2.
+
 `--rule-yield` is a corpus report, not a mark: every live gate rule - the labels
 `hooks/tezgah_gate.py` passes to `_deny`, read from the code with `ast`, so a
 removed rule drops out on its own - with its fires over the real ledgers written

package/docs/testing.md

@@ -66,7 +66,9 @@ conclude from either.
 `.github/workflows/ci.yml` runs on a push to `main` and on a pull request, so
 a branch commit runs once, through its pull request, not twice. A tag push does
 not start it. `release.yml` calls it (`ci.yml:13`) as its `ci` job, so npm and
-brew publish only after CI passed on the release commit. A newer push to the
+brew publish only after CI passed on the release commit. Its first job,
+`on-main`, fails the release when the tag's commit is not on `origin/main`.
+Then no job builds or uploads anything. A newer push to the
 same ref cancels the run it replaces, except on a release (`concurrency`). Every
 job has a `timeout-minutes`, so a hung test cannot hold a runner for six hours.
 No matrix cancels its other legs when one fails (`fail-fast: false`).

package/hooks/hooks.json

@@ -10,10 +10,10 @@
       { "hooks": [ { "type": "command", "command": "TEZGAH_CORE_IN_FILE=1 \"${TEZGAH_PYTHON:-python3}\" \"${CLAUDE_PLUGIN_ROOT}/hooks/projects-auto-init.py\"", "timeout": 10 } ] }
     ],
     "UserPromptSubmit": [
-      { "hooks": [ { "type": "command", "command": "\"${TEZGAH_PYTHON:-python3}\" \"${CLAUDE_PLUGIN_ROOT}/hooks/projects-auto-init.py\"", "timeout": 10 } ] }
+      { "hooks": [ { "type": "command", "command": "TEZGAH_HOST=claude \"${TEZGAH_PYTHON:-python3}\" \"${CLAUDE_PLUGIN_ROOT}/hooks/projects-auto-init.py\"", "timeout": 10 } ] }
     ],
     "PreToolUse": [
-      { "matcher": "Agent|agent|Task|task|Grep|grep|Bash|bash|PowerShell|pwsh|Shell|shell|subagent|Edit|Write|MultiEdit|NotebookEdit|AskUserQuestion", "hooks": [ { "type": "command", "command": "\"${TEZGAH_PYTHON:-python3}\" \"${CLAUDE_PLUGIN_ROOT}/hooks/projects-pretooluse.py\"", "timeout": 5 } ] },
+      { "matcher": "Agent|agent|Task|task|Grep|grep|Bash|bash|PowerShell|pwsh|Shell|shell|Monitor|monitor|subagent|Edit|Write|MultiEdit|NotebookEdit|AskUserQuestion", "hooks": [ { "type": "command", "command": "\"${TEZGAH_PYTHON:-python3}\" \"${CLAUDE_PLUGIN_ROOT}/hooks/projects-pretooluse.py\"", "timeout": 5 } ] },
       { "matcher": "^mcp__(?:.*[^A-Za-z0-9])?(?:append|cancel|clear|click|close|comment|commit|copy|create|delete|deploy|drag|edit|email|evaluate|exec|execute|fill|insert|install|kill|launch|merge|message|mkdir|move|navigate|notify|patch|pause|post|press|print|publish|push|put|release|remove|rename|replace|reply|resume|review|run|save|send|set|share|skip|start|stop|submit|swipe|tag|tap|terminate|type|uninstall|update|upload|write)(?:[^A-Za-z0-9].*)?$", "hooks": [ { "type": "command", "command": "\"${TEZGAH_PYTHON:-python3}\" \"${CLAUDE_PLUGIN_ROOT}/hooks/projects-pretooluse.py\"", "timeout": 5 } ] }
     ],
     "PostToolUse": [

package/hooks/lang/en.json

@@ -153,11 +153,21 @@
    "research|literature|hypothes(is|es)|experiment(al)?|ablation|hyperparameter|benchmark|survey|paper|dataset|compar\\w*|(a/?b|ab) tests?|user interview\\w*|root[ -]?cause (analys|investigat)\\w*|reference review|lit(erature)?[ -]review|evaluat(e|ing) (whether|the (librar|technique|approach|tool|option|alternative))|measure(?! how (?:long|fast|much time)| the (?:time|duration|latency|speed|runtime))\\w*|post[ -]?mortem|error budget|incident (review|report)|is (this|that|it) (actually )?(true|right)|does (this|that) (claim )?hold"
   ],
   "product": [
+   "(?:like|as) an? pm|pm (?:review|lens|perspective|eye|hat)\\w*|intended[- ]vs\\.?[- ]implemented|opportunity solution tree\\w*",
    "product(?!ion|ivity|ive)\\w*|feature(?!s?[ -](?:flags?|toggles?|gates?|branch\\w*))\\w*|roadmap|backlog|prd|north star|jtbd|retention|churn|onboarding|cohort|funnel|conversion rate|pricing|prioriti[sz]\\w*|user research|user interview\\w*|ab test|a/b test|activations?\\b|churn(ed|ing|s)?|funnels?|cohorts?|drop-?off|roadmaps?|priorit(y|ies)|adoption|conversion (rate|funnel|drop\\w*)|packaging|monetiz\\w*|price\\w*|tier(?!s?[ -]lists?)\\w*|positioning|value proposition|\\bicp\\b|ideal customer profile|persona\\w*|segment(?!(?:ation)?[ -]faults?)\\w*|(customer|user|product|problem) discovery|opportunity (tree|solution|space|score)|\\bnps\\b|net promoter|\\bcsat\\b|satisfaction|(customer|user)[ -]?feedback\\w*|support ticket\\w*|competitor\\w*|competitive (analysis|landscape|teardown|benchmark)\\w*|wizard|crud|form (validation|field|error|label)|data table|step flow|search (dropdown|box)|form validation|user management|(admin|users?) (panel|screen|page|list)",
    {
     "ui_ask": "(?:admin|users?) table"
    }
   ],
+  "design": [
+   "ui(?![ -]threads?\\b)|ux|design (?:system|token|language|guide|review|critique|audit|library)\\w*|wireframe\\w*|mockups?|figma|(?<!memory )(?<!data )(?<!disk )(?<!file )(?<!struct )(?<!record )(?<!keyboard )layout|responsive|(?:ui|react|vue|svelte|angular|frontend|front-end|button|form|card|modal|input|nav\\w*|header|footer|sidebar)[ -]components?|components? librar\\w*|ui kit|style guides?|typography|colou?rs? (?:palette|scheme|token|contrast)\\w*|palette|dark mode|a11y|wcag\\w*|accessibility|usability|visual (?:hierarchy|design|consistency|critique|review)|landing page|(?:design|sketch|mock)\\w* (?:\\w+ ){0,3}(?:dashboard|screen|page|form|view|layout|landing)s?|improve (?:this|the|our) (?:screen|page|view|form|layout|dashboard)s?|click(?:ing)? through (?:\\w+ ){0,2}(?:screens?|pages?|app|ui)",
+   {
+    "ui_ask": "confusing|inconsistent|cluttered|ugly|clunky"
+   },
+   {
+    "ui_ask": "redesign\\w*|restyle\\w*|revamp\\w*|accessible"
+   }
+  ],
   "graph": [
    "who calls|callers?|call sites?|who uses|what breaks|blast radius|where is|where's|definition of|who invokes|impact (analysis|assessment|of (changing|renaming|removing|deleting|editing))|change impact|ripple effect|(who|what|anything)( else)? (depends on|uses)|dependents? of|dependenc(y|ies) (graph|direction|inversion|cycle)|layer (violation|boundar\\w*|direction)|layering|who imports|which (files|modules) import|call (graph|chain|path|tree|hierarchy)|transitive callers?|callers? of|usages? of|find usages|references? to|is \\w+ (still )?(used|referenced|called)|never called|dead code|unused (code|function|method|module|export|class|variable|import|param\\w*)|unreachable code|orphan(ed)? (module|file|code)|circular (import|dependenc\\w*)|(import|dependenc\\w*) cycle|cycle (between|detection)|(code|app|program|service|main|server|cli|boot) entry ?points?|boot (path|sequence|order)|startup (path|sequence|order)|how is (this|it|\\w+) wired|wired up|rename\\w*"
   ]
@@ -214,6 +224,9 @@
   "plan_build": [
    "\\bplans?\\b[^.?!\\n]{0,40}\\b(?:is ready|ready|written|drafted)\\b|\\b(?:wrote|opened|drafted|added|created) (?:the |a |this )?plans?\\b|\\b(?:build|implement|execute|start on|start) (?:the |this |that )?plans?\\b"
   ],
+  "owner_choice": [
+   "\\bwhich (?:variant|option|approach|design)s?\\b|\\b(?:v\\d+|option [a-h1-9]|variant [a-z0-9]+),? or (?:v\\d+|option [a-h1-9]|variant [a-z0-9]+)\\b|\\b(?:another|other|the other) session\\w*|\\bsession'?s (?:line|plan|branch)\\b|\\bclos\\w*\\b[^?\\n]{0,40}\\bline\\b|--allow-open|--limit\\b|\\.staging\\b|\\bhidden (?:dir|directory|folder)\\b|\\b(?:another|a different) (?:folder|dir|directory)\\b|\\b(?:work ?)?around\\b[^?\\n]{0,40}\\brefusal\\b"
+  ],
   "spend": [
    "spend\\w*|\\bpaid\\b|billing|\\bcost\\w*|\\bbudget\\w*|\\busd\\b|\\$\\s?\\d"
   ],
@@ -226,6 +239,12 @@
   "cheap_answer": [
    "yes|yep|yeah|ok(?:ay)?|sure|go(?: on| ahead)?|continue|proceed|do it|carry on|keep going|approved"
   ],
+  "hand_over": [
+   "\\bplease\\s+(?:run|execute)\\b|\\byour\\s+(?:step|turn|move)\\b(?=\\s*[:.;,!]|\\s*$)|\\bover to you\\b"
+  ],
+  "ran_it": [
+   "\\bi (?:ran|tested|verified|checked|executed)\\b|\\bran the (?:tests?|suite|build|lint|checks?)\\b|\\bwe ran\\b"
+  ],
   "outward": {
    "push": "\\bpush\\w*",
    "release": "\\brelease\\w*|\\btag(?:s|ged|ging)?\\b|version bump|bump the version",
@@ -319,6 +338,8 @@
   "slice_lead": "",
   "ready_lead": "",
   "ask_lead": "",
+  "handed_lead": "",
+  "plan_lead": "",
   "approval_heading": "Approval questions",
   "recommended": "Recommended",
   "grant_examples": "\"push without asking\", \"I approve\""

package/hooks/lang/tr.json

@@ -196,7 +196,17 @@
    "araştır\\w*|literatür\\w*|hipotez\\w*|deney|karşılaştır\\w*|karşılaştırmalı analiz|kıyaslama\\w*|varyant\\w*|a/b testi|(kullanıcı|müşteri) (interview\\w*|görüşme\\w*|mülakat\\w*)|user (görüşme\\w*|mülakat\\w*)|anket\\w*|kök neden (analiz|araştır)\\w*|kaynak tarama\\w*|referans tarama\\w*|(yöntem|yaklaşım|kütüphane|varyant|seçenek)\\w* karşılaştır\\w*|ölçüm\\w*|deneysel\\w*|deneyler\\w*|makale\\w*|veri (seti|kümesi)|olay sonrası (analiz|değerlendirme)\\w*|hata bütçe\\w*|gerçekten (doğru|öyle) mu|kanıt\\w*|test kapsam\\w*|kapsam oran\\w*|kapsamı (artır|yükselt)\\w*|doküman(tasyon)?\\w* (yetersiz|eksik|kötü|zayıf)"
   ],
   "product": [
-   "ürün\\w*|yol harita\\w*|kuzey yıldız\\w*|aktivasyon\\w*|dönüşüm\\w*|fiyatlandır\\w*|önceliklendir\\w*|kullanıcı araştırma\\w*|ürün keşf\\w*|müşteri geri bildirim\\w*|benimsen\\w*|konumlandır\\w*|segmentasyon\\w*|hedef kitle\\w*|(kullanıcı|müşteri) discovery|keşif (görüşme|çalışma)\\w*|fırsat\\w*|memnuniyet\\w*|anket\\w*|(kullanıcı|müşteri)[ -]?(feedback|geri bildirim)\\w*|(customer|user)[ -]?geri bildirim\\w*|şikayet\\w*|rakip\\w*|rekabet\\w*|pazar pay\\w*|terk oran\\w*|elde tutma\\w*|abonelik\\w*|gelir model\\w*|özellik\\w*|sayfa(?! numara)\\w*|ekran(?! kart)(?!a\\b[^.!?\\n]{0,40}\\b(?:yaz|bas)\\w*)\\w*|arayüz\\w*|arama kutu\\w*|filtre\\w*|tablo(?![^.!?\\n]{0,60}\\b(?:sütun|kolon|sql|index|indeks)\\w*)\\w*|adım(?! sayı)\\w*|kullanıcı liste\\w*|form(u|un|da|daki|lar|ları|unu)\\w*"
+   "pm (?:g[öo]z[üu]yle|gibi|olarak|bak[ıi][şs]\\w*|perspektif\\w*|a[çc][ıi]s[ıi]ndan)|urun(?!\\w*_)\\w*|[öo]zellik\\w*|[öo]zelli[ğg]\\w*[^.!?\\n]{0,40}\\b(?:denetle|incele|analiz|de[ğg]erlendir|g[öo]zden ge[çc]ir)\\w*|[öo]nceliklendir\\w*|(?:[öo]zellik|[üu]r[üu]n|backlog|roadmap|yol harita|talep|istek)\\w* (?:\\w+ ){0,2}[öo]ncelik\\w*|[öo]ncelik\\w* (?:\\w+ ){0,2}(?:[öo]zellik|[üu]r[üu]n|backlog|roadmap|yol harita|talep|istek)\\w*|de[ğg]erlendir\\w* (?:\\w+ ){0,2}(?:uygulama|[üu]r[üu]n|[öo]zellik)\\w*|(?:uygulama|[üu]r[üu]n|[öo]zellik)\\w* (?:\\w+ ){0,2}de[ğg]erlendir\\w*",
+   "ürün(?!\\w*_)\\w*|yol harita\\w*|kuzey yıldız\\w*|aktivasyon\\w*|dönüşüm\\w*|fiyatlandır\\w*|önceliklendir\\w*|kullanıcı araştırma\\w*|ürün keşf\\w*|müşteri geri bildirim\\w*|benimsen\\w*|konumlandır\\w*|segmentasyon\\w*|hedef kitle\\w*|(kullanıcı|müşteri) discovery|keşif (görüşme|çalışma)\\w*|fırsat\\w*|memnuniyet\\w*|anket\\w*|(kullanıcı|müşteri)[ -]?(feedback|geri bildirim)\\w*|(customer|user)[ -]?geri bildirim\\w*|şikayet\\w*|rakip\\w*|rekabet\\w*|pazar pay\\w*|terk oran\\w*|elde tutma\\w*|abonelik\\w*|gelir model\\w*|özellik\\w*|sayfa(?! numara)\\w*|ekran(?! kart)(?!a\\b[^.!?\\n]{0,40}\\b(?:yaz|bas)\\w*)\\w*|arayüz\\w*|arama kutu\\w*|filtre\\w*|tablo(?![^.!?\\n]{0,60}\\b(?:sütun|kolon|sql|index|indeks)\\w*)\\w*|adım(?! sayı)\\w*|kullanıcı liste\\w*|form(u|un|da|daki|lar|ları|unu)\\w*[^.!?\\n]{0,40}\\b(?:incele|denetle|analiz|de[ğg]erlendir|iyile[şs]tir|sadele[şs]tir|kontrol)\\w*"
+  ],
+  "design": [
+   "(?:(?<![\\w'’] )|(?<=\\bbu )|(?<=\\bşu ))tasar[ıi]m\\w*|(?:dashboard|pano|g[öo]sterge panel|ekran|sayfa|aray[üu]z|landing|logo|ikon)\\w* (?:\\w+ ){0,2}(?:tasarla|tasar[ıi]m|redesign)\\w*|(?:web[ -]?)?site\\w* tasar[ıi]m\\w*|(?:mobil |web )?uygulama\\w* tasar[ıi]m\\w*|(?:web|mobil) tasar[ıi]m\\w*|tasar[ıi]m (?:sistem|token|dil|rehber)\\w*|yeniden tasarla\\w*|ba[şs]tan tasarla\\w*|design [çc]al[ıi][şs]ma\\w*|aray[üu]z\\w*|(?:ui|aray[üu]z|form|buton|d[üu][ğg]me|kart|men[üu]|sayfa|ekran|react|vue)\\w* (?:\\w+ )?bile[şs]en\\w*|bile[şs]en (?:k[üu]t[üu]phane|sistem)\\w*|tipografi\\w*|renk\\w*|tema(?:s[ıi]|y[ıi]|lar\\w*)?|kullan[ıi]labilirlik\\w*|eri[şs]ilebilirli[kğg]\\w*|g[öo]rsel (?:hiyerar[şs]i|tutarl[ıi]l[ıi]k|denetim|ele[şs]tiri)\\w*",
+   {
+    "ui_ask": "iyile[şs]tir\\w*|g[üu]zelle[şs]tir\\w*|d[üu]zenle\\w*|kar[ıi][şs][ıi]k\\w*|tutars[ıi]z\\w*|da[ğg][ıi]n[ıi]k\\w*|[çc]irkin\\w*"
+   },
+   {
+    "ui_ask": "eri[şs]ilebilir\\w*"
+   }
   ],
   "graph": [
    "kim çağır\\w*|çağrı yerleri|nerede tanımlı|nasıl bağlan\\w*|etkilenir\\w*|hangi dosyalar etkilen\\w*|etki (analiz|alan)\\w*|neyi etkile\\w*|neleri etkile\\w*|etkilen\\w*|bağımlı\\w*|kullanılmayan\\w*|kullanılmıyor|ölü kod|döngüsel (bağımlılık|import)\\w*|giriş nokta\\w*|uygulama (nerede|nereden) başl\\w*|kim import ed\\w*|(fonksiyon|metod|sınıf|modül|api)\\w* (kim|nerede) kullan\\w*|nerede (kullanıl|çağrıl)\\w*|tanım\\w* nerede|kullanım yerleri|çağrı (zinciri|grafiği)|bağımlılık (grafiği|ağacı)|yeniden adlandır\\w*"
@@ -258,6 +268,9 @@
   "plan_build": [
    "\\bplan\\w*[^.?!\\n]{0,40}(?:hazır|yazıldı|yazdım|oluşturdum|oluşturuldu|açtım|açıldı|eklendi|uygula\\w*|başla\\w*|kur\\w*)|(?:yazdığım|açtığım|hazırladığım|oluşturduğum) plan\\w*"
   ],
+  "owner_choice": [
+   "\\b(?:v\\d+|seçenek \\w+|varyant \\w+)(?:'\\w+)? m[ıiuü]\\b,? (?:v\\d+|(?:seçenek |varyant )?\\w+)(?:'\\w+)? m[ıiuü]\\b|\\bhangi (?:varyant|seçenek|opsiyon|yaklaşım|tasarım)\\w*|\\bhangisini seç\\w*|\\b(?:başka|diğer) (?:bir )?oturum\\w*|\\bhatt[ıi]\\w* kapat\\w*|--allow-open|--limit\\b|\\.staging\\b|\\bgizli (?:dizin|klasör)\\w*|\\b(?:başka|farklı) (?:bir )?(?:klasör|dizin)\\w*|\\bred\\w* (?:aş|etrafından)\\w*|\\bengel\\w* (?:aş|etrafından)\\w*"
+  ],
   "spend": [
    "ücret\\w*|ödeme\\w*|harca\\w*|bütçe\\w*|maliyet\\w*|\\bpara\\b|dolar\\w*|\\$\\s?\\d"
   ],
@@ -270,6 +283,12 @@
   "cheap_answer": [
    "evet|tamam|olur|devam(?: et)?|yap|başla|ilerle|okey|onay(?:lıyorum)?"
   ],
+  "hand_over": [
+   "\\b(?:sen|siz)\\b[^.!?\\n]{0,32}?\\bçalıştır(?:ın|ınız)?\\b|\\bsenin adımın\\b|\\badım(?:ı)? sende\\b|\\bsıradaki adım(?:ı)? sen\\b"
+  ],
+  "ran_it": [
+   "\\b(?:test(?:ler)?i|suite'?i|komutu|kontrolü|denetimi)?\\s*(?:çalıştırdım|koştum|çalıştırdık)\\b|\\btest ettim\\b|\\bdoğruladım\\b|\\bkontrol ettim\\b|\\bdenetledim\\b"
+  ],
   "outward": {
    "push": "\\bpush\\w*",
    "release": "versiyon\\w*|sürüm\\w*|\\brelease\\w*|etiket\\w*",
@@ -340,6 +359,8 @@
   "slice_lead": "Durmuş bir dilim var. ",
   "ready_lead": "Hazır adımlar bekliyor. ",
   "ask_lead": "Bu soruyu sorma: kararı ver ya da önerili onay sorusuna çevir. ",
+  "handed_lead": "Bu adımı sen çalıştır, kullanıcıya bırakma. ",
+  "plan_lead": "Plan modunda hiçbir şey koşmadı. ",
   "approval_heading": "Onay için sorular",
   "recommended": "Öneri",
   "grant_examples": "\"bana sormadan pushla\", \"onay veriyorum\""

package/hooks/projects-stop.py

@@ -64,7 +64,8 @@ def main():
     # a core that cannot answer has refused nothing, so the turn ends: the rule
     # never traps a session, and that holds for its own failure too
     reason = safe(session_id, stop_reason, p.get("last_assistant_message"),
-                  session_id, cwd=cwd, after_block=bool(p.get("stop_hook_active")))
+                  session_id, cwd=cwd, after_block=bool(p.get("stop_hook_active")),
+                  mode=p.get("permission_mode"))
     if reason:
         json.dump({"decision": "block", "reason": reason}, sys.stdout)