@jjanczur/tyran 0.1.0 → 0.1.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -11,7 +11,7 @@
11
11
  {
12
12
  "name": "tyran",
13
13
  "description": "A task conductor for Claude Code: multi-agent orchestration with an enforced evidence contract, repo-specific learning, and update-safe local evolution.",
14
- "version": "0.1.0",
14
+ "version": "0.1.2",
15
15
  "source": "./",
16
16
  "author": {
17
17
  "name": "Jacek Janczura",
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "tyran",
3
- "version": "0.1.0",
3
+ "version": "0.1.2",
4
4
  "description": "A task conductor for Claude Code: multi-agent orchestration with an enforced evidence contract, repo-specific learning, and update-safe local evolution.",
5
5
  "author": {
6
6
  "name": "Jacek Janczura",
package/CHANGELOG.md CHANGED
@@ -1,6 +1,169 @@
1
1
  # Changelog
2
2
 
3
- ## 0.1.0 — unreleased
3
+ ## 0.1.2 — 2026-07-29
4
+
5
+ From a field report on a real 0.1.1 install. Every item below is a failure
6
+ somebody hit, not one anybody predicted.
7
+
8
+ ### The worktree with no boundary
9
+
10
+ **`git worktree add` carries tracked files only, and the policy gate is
11
+ deliberately silent in a repo with no `.tyran/`.** Put those together and an
12
+ uncommitted `.tyran/` means every worktree the conductor creates has no config
13
+ and no policy — so the agents there run with **no autonomy class and no path
14
+ classes at all**. Nothing is refused, nothing fails, and `git push origin main`
15
+ returns `PASS` on line 1213 of the gate. Measured: four worktrees, four ungated
16
+ implementers, and no way for the operator to see it.
17
+
18
+ This is the worst kind of defect this project can have, because Tyran's whole
19
+ argument is that the boundary is a mechanism rather than a promise — and here
20
+ it was absent precisely where the most agents run.
21
+
22
+ Three changes, none of which touch the gate:
23
+
24
+ - `doctor --state` gains **`tyran-dir-untracked`** (warning): git is asked
25
+ whether anything under `.tyran/` is tracked. It returns *no answer* outside a
26
+ work tree rather than inventing one, the same discipline the lockfile rule
27
+ below uses;
28
+ - `/tyran:setup` now ends by asking for the commit, **with the reason**, rather
29
+ than leaving `.tyran/` sitting untracked;
30
+ - the conductor treats a worktree without `.tyran/` as a precondition failure,
31
+ and is told to fix it by committing on the main checkout — not by copying the
32
+ file in, which just makes four divergent configs one layer down.
33
+
34
+ ### Two inferences that produced a broken gate
35
+
36
+ **A validation command that watches does not fail an agent, it hangs it.**
37
+ Setup wrote `pnpm test` into a repo whose `test` script is bare `vitest` —
38
+ watch mode, no output, no timeout, a session that simply stops. `detectValidation`
39
+ now reroutes to a run-once variant (`test:run`, `test:ci`, `test:once`) and, if
40
+ there is none, **leaves the command out** and flags the config rather than
41
+ writing a booby trap. The rule is deliberately narrow — `jest` runs once by
42
+ default and is not treated as a watcher — because a false positive here silently
43
+ drops a real test command.
44
+
45
+ **A lockfile on disk is not evidence that it is the repo's lockfile.** The same
46
+ install had `pnpm-lock.yaml` gitignored and untracked alongside the tracked
47
+ `package-lock.json` that its deploy actually builds from, and `.gitignore` said
48
+ so in words. Picking by disk order chose pnpm, and then every single validation
49
+ command was wrong. `detectPackageManager` now asks git which lockfiles are
50
+ tracked and says which one it rejected. An empty `git ls-files` means git could
51
+ not answer and changes nothing — "not a git repo" is not evidence that a file is
52
+ ignored.
53
+
54
+ Both replay clean against the repository that produced them.
55
+
56
+ ### Errors that named the wrong cause
57
+
58
+ `yaml-lite` answered every unbalanced quote with *"quote the whole value if it
59
+ contains an apostrophe"*. An operator writing a long `source:` as a multi-line
60
+ double-quoted string was sent hunting an apostrophe that did not exist, three
61
+ times. A value that opens a quote and never closes it now says exactly that,
62
+ and says the subset has no multi-line scalars. The apostrophe advice survives
63
+ only on the case it is actually about.
64
+
65
+ `doctor.mjs` with no flag printed *"--state is the only mode today"* long after
66
+ `--hooks` shipped, sending readers to look for a flag they already had.
67
+
68
+ The generated `config.yaml` header now states the YAML subset and asks for the
69
+ commit, since both traps are sprung by hand-editing a file whose constraints
70
+ were documented somewhere else.
71
+
72
+ ### Still open, deliberately
73
+
74
+ The report also found that the gate matches on command TEXT, so `ls -1 .npmrc`
75
+ is refused as a credential read though it reads nothing, and that a `GATED`
76
+ refusal tells an agent to write a diff for the operator while `outside this
77
+ repository is never autonomous` refuses the scratch directory that is the
78
+ obvious place to put it. That second one is self-defeating and worth fixing.
79
+ Both are changes to refusal semantics in a KERNEL file and are not being made
80
+ in the same pass as everything above.
81
+
82
+ ## 0.1.1 — 2026-07-29
83
+
84
+ Both entries below come from one install of 0.1.0 on a real repository. The
85
+ version is bumped rather than folded into 0.1.0 because `claude plugin update`
86
+ compares **versions**, not commits: with the manifest left at 0.1.0 it reports
87
+ "already at the latest version" and users never receive the fix.
88
+
89
+ ### `.tyran/config.yaml` is AUTO, and what that costs is written down
90
+
91
+ **The file that says how to validate this repo was one an agent could not
92
+ fix.** Measured on the same install as below: setup inferred `pnpm test`,
93
+ which in that repo is bare `vitest` — watch mode, it never exits, and it would
94
+ have hung every agent handed to it. The session that *discovered* that could
95
+ not repair the file that said it. It produced a diff, then a heredoc for the
96
+ operator to paste, then explained why it would not route around its own gate.
97
+ It was right not to. The gate was wrong.
98
+
99
+ So the shipped policy classes `.tyran/config.yaml` `AUTO`. It is mostly a
100
+ description of the repository — package manager, validation commands, shared
101
+ zones — and the agent that finds it wrong is what is best placed to correct it.
102
+
103
+ The cost is real and is now stated in all four places a user might read
104
+ (`templates/policies/autonomy.yaml`, `docs/configuration.md`,
105
+ `docs/policy-gate.md`, `docs/self-improvement.md`), because an unstated cost is
106
+ a false guarantee: **`autonomy:` lives in this file too, so an agent can raise
107
+ its own deployment class from P1 to P3 and then push to main.** What remains is
108
+ weaker and is said as such — nothing *infers* a raise, and a raise is a diff in
109
+ a committed file with its provenance beside it. Note that `docs/policy-gate.md`
110
+ had already measured this exact escalation happening *under* GATED, in an
111
+ unattended main loop, wherever `Write` is allow-listed; GATED was buying less
112
+ here than it looked like it was buying.
113
+
114
+ The push gate's own refusal named the old class too, and naming a class there
115
+ was the underlying mistake rather than naming the wrong one: the text is
116
+ per-repo configuration, so any class it prints goes stale the moment somebody
117
+ reclassifies the file. It now says the policy decides and names nothing, and
118
+ the test asserts the *absence* of a class claim instead of pinning a string
119
+ that has already been wrong once.
120
+
121
+ Two claims that had gone false are corrected rather than left standing:
122
+ `README.md`'s "the config file holding it is GATED, not KERNEL" and
123
+ `docs/configuration.md`'s "Tyran never raises it on its own — that path is
124
+ blocked by the policy hook". A test now pins the class *and* pins that the
125
+ template still spells out what it gives up, because flipping it back is a
126
+ one-word edit nobody would notice in review.
127
+
128
+ ### Setup no longer locks the repository it is setting up
129
+
130
+ **`/tyran:setup` created `.tyran/config.yaml` and stopped, and that one
131
+ missing file made the repository unwritable.** The policy gate is silent in a
132
+ repo with no `.tyran/` directory and refuses every write in a repo that has
133
+ one without `.tyran/policies/autonomy.yaml` under it. Both halves are right;
134
+ their seam was not. Setup's own first command moved a fresh repo from
135
+ "unmanaged" to "refuses everything" — including the write that would have
136
+ installed the missing policy. Measured on a real install: the session ended
137
+ with the operator being handed a `mkdir` and a `cp` to run by hand, in the
138
+ middle of a one-command setup.
139
+
140
+ `scan-repo.mjs` now installs the policy from the shipped template **before**
141
+ it writes the config, and removes what it created if it cannot, so a failed
142
+ bootstrap cannot leave behind the exact state it exists to prevent. A repo
143
+ from before this change is repaired with `--ensure-policy`, which touches
144
+ nothing else.
145
+
146
+ That is bootstrap, not a loop authorizing itself, and the difference is
147
+ mechanical rather than a matter of intent: it only ever creates — an existing
148
+ policy is never read, merged or overwritten — and what it writes is the
149
+ shipped template byte for byte, the strictest default Tyran has. No input
150
+ makes it emit something weaker. Editing the file afterwards is human-only,
151
+ enforced exactly as before. Three tests pin the properties that make that
152
+ sentence true rather than aspirational: byte-identity with the template, an
153
+ untouched hand-written policy across setup and repair, and no `.tyran/` left
154
+ on disk when the install fails.
155
+
156
+ Two smaller things the same incident exposed. `doctor`'s `policy-missing`
157
+ finding was severity `info` — a repository where every tool call is denied,
158
+ reported as a note — and its suggested fix was a `cp` naming the policy path,
159
+ which the gate refuses when an agent runs it; it is now an `error` carrying a
160
+ command that works. And the documented `schema.mjs validate policy
161
+ .tyran/policies/…` line is one a **human** runs: inside a session it names a
162
+ path under `.tyran/policies/**` and is refused like any other. `doctor --state
163
+ --dir .tyran` validates the same file and names only the directory. The docs
164
+ say so now, and `/tyran:setup` runs the latter.
165
+
166
+ ## 0.1.0 — 2026-07-27
4
167
 
5
168
  ### Six protocol skills, and the budget raised once to pay for them
6
169
 
package/README.md CHANGED
@@ -21,11 +21,15 @@
21
21
  <h3 align="center">The more you use it, the better it gets.</h3>
22
22
 
23
23
  <p align="center">
24
- Tyran is a task conductor for Claude Code: it interviews you, plans, and drives a team of
25
- fresh-context agents through your work — <b>refusing to believe any agent that can't show
26
- raw command output as proof</b>, routing every subtask to the cheapest model that can do
27
- it, and keeping each agent in the context band where it still thinks clearly. When the
28
- work is done, it will not let the initiative end until it has learned something from it.
24
+ Never written a skill, spun up a subagent, or run an agent team yourself? Tyran does all
25
+ three automatically, on your behalf — and the retrospective it runs after every initiative
26
+ means it keeps learning your repo's way of working instead of repeating a generic playbook.
27
+ </p>
28
+
29
+ <p align="center">
30
+ Under the hood: it interviews you, plans, and drives a team of fresh-context agents through
31
+ your work, routing every subtask to the cheapest model that can actually do it — so a file
32
+ rename doesn't burn through your subscription at top-tier rates.
29
33
  </p>
30
34
 
31
35
  ---
@@ -142,7 +146,7 @@ Initiative 20 Has written repo-specific skills for your recurring work and
142
146
  ```text
143
147
  /plugin marketplace add jjanczur/tyran
144
148
  /plugin install tyran@tyran
145
- /tyran:setup # scans the repo, writes .tyran/config.yaml, installs /tyran
149
+ /tyran:setup # scans the repo, writes .tyran/, installs /tyran
146
150
  /tyran # then just describe what you want done
147
151
  ```
148
152
 
@@ -158,9 +162,11 @@ Install the Tyran plugin in this repository and set it up for me.
158
162
  1. Run: claude plugin marketplace add jjanczur/tyran
159
163
  2. Run: claude plugin install tyran@tyran
160
164
  3. Tell me to restart Claude Code, so the hooks and agents load.
161
- 4. After the restart, run /tyran:setup. It scans this repo and writes
162
- .tyran/config.yaml. Walk me through what it inferred - especially the
163
- validation commands and the deployment autonomy class - before we commit it.
165
+ 4. After the restart, run /tyran:setup, then /tyran:doctor. Setup writes the
166
+ two files under .tyran/ that Tyran needs - the config, and the autonomy
167
+ policy the write gate reads - and doctor proves the install actually
168
+ works. Walk me through what setup inferred - especially the validation
169
+ commands and the deployment autonomy class - before we commit it.
164
170
 
165
171
  Docs: https://jjanczur.github.io/tyran/getting-started/
166
172
  ```
@@ -278,10 +284,12 @@ echo "wrong branch — hold everything" > .tyran/STOP
278
284
  force-pushes are refused; a push to your production branch is refused at the
279
285
  strictest autonomy class, including through `HEAD`, `@`, a shell variable or
280
286
  a `git -c` alias. **What it does not do is stop an agent from raising the
281
- class itself:** the config file holding it is GATED, not KERNEL, so an agent
282
- with a broad allow-list can edit it in the main loop — measured, not
283
- supposed. Autonomy classes are a convention the gate enforces downward, not
284
- a lock. Not a firewall, and
287
+ class itself:** the config file holding it is `AUTO` in the shipped policy,
288
+ so an agent can edit it. That is deliberate — the same file holds your
289
+ validation commands, and an agent that cannot correct a wrong one is an
290
+ agent that runs your test suite wrong forever. Set the rule back to `GATED`
291
+ in one word if you want the other trade. Autonomy classes are a convention
292
+ the gate enforces downward, not a lock. Not a firewall, and
285
293
  [docs/hooks.md](docs/hooks.md) says exactly where it stops — including the
286
294
  scanner's own measured false-negative rate.
287
295
 
@@ -341,53 +349,48 @@ one directory each.</sub>
341
349
 
342
350
  ## Command-line use (outside Claude Code)
343
351
 
344
- A single `bin/tyran.mjs` entry point exposes the same scripts the plugin's
345
- hooks and skills already call — `doctor`, `scan-repo`, `tiers`, `journal`,
346
- `schema`, `stop-check`, `scan-control-chars`, `desc-budget` — for a shell or a
347
- CI job with no Claude Code session. Zero dependencies; `--help` lists them.
352
+ Published on npm as [`@jjanczur/tyran`](https://www.npmjs.com/package/@jjanczur/tyran)
353
+ — scoped, not the bare name `tyran`, which carries an unpublish tombstone from
354
+ 2021-03-30 that npm's abuse policy blocks anyone from reusing, permanently. The
355
+ command stays `tyran` either way; only the registry name is scoped, the same
356
+ pattern as `@angular/cli` giving you `ng`. `bin/tyran.mjs` exposes the same
357
+ scripts the plugin's hooks and skills already call — `doctor`, `scan-repo`,
358
+ `tiers`, `journal`, `schema`, `stop-check`, `scan-control-chars`,
359
+ `desc-budget` — for a shell or a CI job with no Claude Code session. Zero
360
+ dependencies; `--help` lists them.
348
361
 
349
362
  ```bash
350
- node bin/tyran.mjs doctor --hooks # is any gate installed but unable to fire?
351
- node bin/tyran.mjs scan-repo --dir . # what this repo looks like, with provenance
363
+ npx @jjanczur/tyran doctor --hooks # is any gate installed but unable to fire?
364
+ npx @jjanczur/tyran scan-repo --dir . # what this repo looks like, with provenance
352
365
  ```
353
366
 
354
367
  Exit codes propagate to the digit, so a CI step reddens exactly when the
355
368
  underlying script does.
356
369
 
357
- > **Not on npm yet — so `npx @jjanczur/tyran` does not work today.** The
358
- > package is built, versioned and tested; only the publish is outstanding.
359
- > Saying it works here before that is true would be exactly the kind of claim
360
- > this project refuses to accept from its own agents. It publishes under the
361
- > scope, not the bare name `tyran` — that name carries an npm unpublish
362
- > tombstone from 2021-03-30, and npm's policy against reusing an unpublished
363
- > name is permanent, for anyone, not a grace period this project could wait
364
- > out. The command stays `tyran` either way; only the registry name is scoped.
365
- >
366
- > **And it would not install the Claude Code plugin either.** For that, run
370
+ > **This does not install the Claude Code plugin.** For that, run
367
371
  > `/plugin marketplace add jjanczur/tyran` inside Claude Code — that name is
368
- > the Claude Code marketplace identity, a separate namespace from npm and
369
- > unaffected by any of the above. The npm package is the tooling; the plugin
370
- > is the conductor.
372
+ > the Claude Code marketplace identity, a separate namespace from npm. The npm
373
+ > package is the tooling; the plugin is the conductor.
371
374
 
372
375
  ## Documentation
373
376
 
374
377
  **Everything below also reads as a site: [jjanczur.github.io/tyran](https://jjanczur.github.io/tyran/)**
375
378
  — same text, with search, rendered diagrams and per-page status badges.
376
379
 
377
- - 📖 [Getting started](docs/getting-started.md)
378
- - 🧰 [Skills and agents](docs/skills.md) — all fourteen skills and four agents, what each assumes, and who invokes it
379
- - ⚙️ [Configuration](docs/configuration.md) — `.tyran/config.yaml`, cost profiles, autonomy classes
380
- - 🎭 [The roster and model routing](docs/agents.md) — the four agents, the tier and effort table, dynamic overrides, the `.tyran/STOP` brake (shipped)
381
- - 🏛️ [Architecture](docs/architecture.md) — the three layers, the journal, the hooks
382
- - 📜 [Journal reference](docs/journal.md) — the append-only event schema (shipped)
383
- - 🧾 [Projections](docs/projections.md) — generated `STATE.md` / `PROGRESS.md` and `--check` (shipped)
384
- - 🩺 [Doctor](docs/doctor.md) — `--state` consistency check: drift, orphan leases, dead policy rules (shipped)
385
- - 🪝 [Hook runtime](docs/hooks.md) — gates vs probes, why hooks fail open, and what the deadline really promises (shipped)
386
- - 🧾 [Evidence gate](docs/evidence-gate.md) — the criterion, who it binds, the recorded escape hatch, and the line between silence and forgery (shipped)
387
- - 🛡️ [Policy gate](docs/policy-gate.md) — path classes, the deployment class, the one rule on reads, and where each stops (shipped)
388
- - 🧠 [Self-improvement](docs/self-improvement.md) — how Tyran learns your repo, and its guardrails
389
- - ❓ [FAQ](docs/faq.md)
390
- - 🤝 [Contributing](CONTRIBUTING.md)
380
+ - 📖 [Getting started](https://jjanczur.github.io/tyran/getting-started/)
381
+ - 🧰 [Skills and agents](https://jjanczur.github.io/tyran/skills/) — all fourteen skills and four agents, what each assumes, and who invokes it
382
+ - ⚙️ [Configuration](https://jjanczur.github.io/tyran/configuration/) — `.tyran/config.yaml`, cost profiles, autonomy classes
383
+ - 🎭 [The roster and model routing](https://jjanczur.github.io/tyran/agents/) — the four agents, the tier and effort table, dynamic overrides, the `.tyran/STOP` brake (shipped)
384
+ - 🏛️ [Architecture](https://jjanczur.github.io/tyran/architecture/) — the three layers, the journal, the hooks
385
+ - 📜 [Journal reference](https://jjanczur.github.io/tyran/journal/) — the append-only event schema (shipped)
386
+ - 🧾 [Projections](https://jjanczur.github.io/tyran/projections/) — generated `STATE.md` / `PROGRESS.md` and `--check` (shipped)
387
+ - 🩺 [Doctor](https://jjanczur.github.io/tyran/doctor/) — `--state` consistency check: drift, orphan leases, dead policy rules (shipped)
388
+ - 🪝 [Hook runtime](https://jjanczur.github.io/tyran/hooks/) — gates vs probes, why hooks fail open, and what the deadline really promises (shipped)
389
+ - 🧾 [Evidence gate](https://jjanczur.github.io/tyran/evidence-gate/) — the criterion, who it binds, the recorded escape hatch, and the line between silence and forgery (shipped)
390
+ - 🛡️ [Policy gate](https://jjanczur.github.io/tyran/policy-gate/) — path classes, the deployment class, the one rule on reads, and where each stops (shipped)
391
+ - 🧠 [Self-improvement](https://jjanczur.github.io/tyran/self-improvement/) — how Tyran learns your repo, and its guardrails
392
+ - ❓ [FAQ](https://jjanczur.github.io/tyran/faq/)
393
+ - 🤝 [Contributing](CONTRIBUTING.md) — not a docs-site page, so this one stays a repo link
391
394
 
392
395
  ## Strategic principles
393
396
 
@@ -406,9 +409,9 @@ underlying script does.
406
409
  generalizable improvements travel to the core as pull requests.
407
410
  6. **Autonomy earns trust in layers.** Detected once, confirmed with you, and
408
411
  enforced downward by the policy gate. Not *"never self-escalated"* — the
409
- file holding the class is GATED rather than KERNEL, and an agent with a
410
- broad allow-list can edit it. Measured, not assumed; see
411
- [the policy gate](docs/policy-gate.md).
412
+ file holding the class is `AUTO`, so an agent can edit it, chosen so that
413
+ an agent can also repair the validation commands stored beside it.
414
+ Measured, not assumed; see [the policy gate](docs/policy-gate.md).
412
415
 
413
416
  ## Roadmap
414
417
 
@@ -1342,8 +1342,11 @@ const DEPLOY_REMEDY = Object.freeze(Object.assign(Object.create(null), {
1342
1342
  'recoverable from this session, which is the whole reason they are outside every autonomy class.',
1343
1343
  production:
1344
1344
  'What to do instead: push to your own branch and open a pull request. The class lives in ' +
1345
- `\`${CONFIG_PATH}\`, which ADR-06 classifies GATED rather than KERNEL, so raising it is an ` +
1346
- 'operator decision by CONVENTION, not by mechanism: measured in review, an agent in an ' +
1345
+ `\`${CONFIG_PATH}\`, and it is your own policy that decides whether an agent may edit that ` +
1346
+ 'file — the shipped template classifies it AUTO — so keeping the class where the operator put ' +
1347
+ 'it is by CONVENTION, not by mechanism. No class is named here on purpose: this text would ' +
1348
+ 'otherwise go stale the moment a repo reclassified the file, and a refusal that states the ' +
1349
+ 'wrong reason is worse than one that states none. Measured in review, an agent in an ' +
1347
1350
  'unattended main loop edited P1 to P3 with no refusal. Treat raising it yourself as the thing ' +
1348
1351
  'you were asked not to do, and say in your report that you wanted to.',
1349
1352
  shared:
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@jjanczur/tyran",
3
- "version": "0.1.0",
3
+ "version": "0.1.2",
4
4
  "description": "Command-line scripts for the Tyran task conductor (doctor, journal, schema, and repo tooling) for CI and terminals outside Claude Code. Installing this package does NOT install the Claude Code plugin — add that separately with `/plugin marketplace add jjanczur/tyran` inside Claude Code.",
5
5
  "type": "module",
6
6
  "engines": {
@@ -49,6 +49,7 @@ import {
49
49
  STATE_FILE,
50
50
  } from './project.mjs';
51
51
  import { classifyPath, normalizePath, validateFile, MANDATORY_KERNEL_PATHS } from './schema.mjs';
52
+ import { gitRunner } from './scan-repo.mjs';
52
53
 
53
54
  /** Severity order is also the report order. `info` never fails the check. */
54
55
  export const SEVERITIES = Object.freeze(['error', 'warning', 'info']);
@@ -175,7 +176,11 @@ export const SEVERITY_BY_CODE = Object.freeze(
175
176
  'knowledge-invalid': 'error',
176
177
  'knowledge-unreadable': 'error',
177
178
  'knowledge-not-a-directory': 'warning',
178
- 'policy-missing': 'info',
179
+ // Not `info`, which is what it was: with `.tyran/` on disk and no policy
180
+ // under it the gate refuses every write in the repository. A severity that
181
+ // does not fail the check reports a locked-out repo as a note.
182
+ 'tyran-dir-untracked': 'warning',
183
+ 'policy-missing': 'error',
179
184
  'policy-invalid': 'error',
180
185
  'policy-unreadable': 'error',
181
186
  'policies-unreadable': 'error',
@@ -980,7 +985,31 @@ function policyFindings(path, repoRoot) {
980
985
  * Run every state check. Pure with respect to time: `now` is either given
981
986
  * or taken from each journal's last event, never from the clock.
982
987
  */
983
- export function runStateChecks({ dir = '.tyran', now = null, staleHours = DEFAULT_STALE_HOURS } = {}) {
988
+ /**
989
+ * Is `.tyran/` committed? The one Tyran failure that is completely silent.
990
+ *
991
+ * `git worktree add` carries TRACKED files only, and the policy gate is
992
+ * deliberately silent in a repository with no `.tyran/` directory. Put those
993
+ * two together and an uncommitted `.tyran/` means every worktree the conductor
994
+ * creates runs with no autonomy class and no path classes — nothing fails,
995
+ * nothing is refused, the boundary is simply absent in the one place the most
996
+ * agents run. Measured on a real install: four worktrees, four ungated
997
+ * implementers, and the operator had no way to see it.
998
+ *
999
+ * Returns null when git cannot answer. The distinction matters exactly as it
1000
+ * does in `detectPackageManager`: "this is not a git repository" is not
1001
+ * evidence that a file is untracked, and reporting it as such would fire on
1002
+ * every temp directory and every fresh `git init`.
1003
+ */
1004
+ export function untrackedTyranDir(repoRoot, run) {
1005
+ if (run(['rev-parse', '--is-inside-work-tree']).trim() !== 'true') return null;
1006
+ // `ls-files` over the directory: any tracked file under it means the tree
1007
+ // travels with a worktree, which is the property being checked. Asking about
1008
+ // config.yaml alone would pass a repo that committed only the policy.
1009
+ return run(['ls-files', '--', '.tyran']).trim() === '';
1010
+ }
1011
+
1012
+ export function runStateChecks({ dir = '.tyran', now = null, staleHours = DEFAULT_STALE_HOURS, run = null } = {}) {
984
1013
  const root = resolve(dir);
985
1014
  if (!existsSync(root)) {
986
1015
  return summarize(dir, [finding('no-state-dir', show(dir), 'no Tyran state directory here — nothing to check')], [
@@ -1009,6 +1038,21 @@ export function runStateChecks({ dir = '.tyran', now = null, staleHours = DEFAUL
1009
1038
  checked.push('config.yaml: absent');
1010
1039
  }
1011
1040
 
1041
+ const untracked = untrackedTyranDir(repoRoot, run ?? gitRunner(repoRoot));
1042
+ if (untracked === true) {
1043
+ findings.push(
1044
+ finding(
1045
+ 'tyran-dir-untracked',
1046
+ show(dir),
1047
+ 'nothing under .tyran/ is tracked by git — `git worktree add` carries tracked files only, and ' +
1048
+ 'the policy gate is silent in a repo with no .tyran/, so every worktree runs with no autonomy ' +
1049
+ 'class and no path classes. Nothing fails; the boundary is absent where the most agents run',
1050
+ `git add ${sq(dir)} && git commit -m 'chore: adopt Tyran'`,
1051
+ ),
1052
+ );
1053
+ }
1054
+ checked.push(`.tyran/ tracked by git: ${untracked === null ? 'unknown (not a git work tree)' : !untracked}`);
1055
+
1012
1056
  const knowledge = yamlFilesIn(join(dir, 'knowledge'), 'knowledge');
1013
1057
  findings.push(...knowledge.findings);
1014
1058
  for (const file of knowledge.files) findings.push(...checkSchemaFile('knowledge', file).findings);
@@ -1022,8 +1066,11 @@ export function runStateChecks({ dir = '.tyran', now = null, staleHours = DEFAUL
1022
1066
  finding(
1023
1067
  'policy-missing',
1024
1068
  show(join(dir, 'policies')),
1025
- 'no autonomy policy — the self-improvement boundary (AUTO / GATED / KERNEL) is undefined for this repo',
1026
- `mkdir -p ${sq(join(dir, 'policies'))} && cp templates/policies/autonomy.yaml ${sq(join(dir, 'policies', 'autonomy.yaml'))}`,
1069
+ 'no autonomy policy, in a repo that HAS a .tyran/ directory — the policy gate fails closed on ' +
1070
+ 'exactly this state, so every write in every session is refused until the file exists. It is not ' +
1071
+ 'a missing nicety; the repository is unusable',
1072
+ `node scripts/scan-repo.mjs --dir ${sq(repoRoot)} --ensure-policy # installs the shipped template; ` +
1073
+ 'never overwrites a policy you wrote',
1027
1074
  ),
1028
1075
  );
1029
1076
  }
@@ -1218,8 +1265,13 @@ function main() {
1218
1265
  if (err instanceof UsageError) {
1219
1266
  console.error(
1220
1267
  'usage: doctor.mjs --state [--dir <.tyran>] [--json] [--now <iso>] [--stale-hours <n>]\n' +
1221
- ' --state is the only mode today; it is required so that adding --env/--config later\n' +
1222
- ' cannot change what a bare invocation means.',
1268
+ ' doctor.mjs --hooks [--plugin-root <dir>] [--json]\n' +
1269
+ '\n' +
1270
+ ' A mode is REQUIRED: `--state` asks whether the record of the work is\n' +
1271
+ ' consistent, `--hooks` whether the gates can fire at all. They answer\n' +
1272
+ ' different questions, and a bare invocation would have to pick one\n' +
1273
+ ' silently. This text said `--state is the only mode today` long after\n' +
1274
+ ' `--hooks` shipped, which sent readers looking for a flag they had.',
1223
1275
  );
1224
1276
  process.exit(2);
1225
1277
  }
@@ -25,13 +25,15 @@
25
25
  *
26
26
  * CLI:
27
27
  * node scan-repo.mjs [--dir <repo>] [--write <path>] [--json]
28
+ * node scan-repo.mjs [--dir <repo>] --ensure-policy
28
29
  * Exit: 0 scanned · 2 usage/IO error
29
30
  */
30
- import { readFileSync, existsSync, writeFileSync, mkdirSync, realpathSync } from 'node:fs';
31
+ import { readFileSync, existsSync, writeFileSync, mkdirSync, rmdirSync, realpathSync } from 'node:fs';
31
32
  import { execFileSync } from 'node:child_process';
32
- import { join, resolve, dirname, extname } from 'node:path';
33
+ import { join, resolve, dirname, extname, relative } from 'node:path';
33
34
  import { fileURLToPath } from 'node:url';
34
- import { stringify } from './yaml-lite.mjs';
35
+ import { parse, stringify } from './yaml-lite.mjs';
36
+ import { validatePolicy } from './schema.mjs';
35
37
  import { escapeInvisible } from './invisible.mjs';
36
38
 
37
39
  /** Lockfile -> package manager. Order matters only for reporting stability. */
@@ -90,9 +92,57 @@ export function gitRunner(dir) {
90
92
  };
91
93
  }
92
94
 
93
- export function detectPackageManager(dir) {
94
- for (const [file, manager] of Object.entries(LOCKFILES)) {
95
- if (existsSync(join(dir, file))) return provenance(manager, `lockfile: ${file}`, 0.95);
95
+ /** The repo's tracked files as a Set, or null when git could not answer. */
96
+ export function trackedFiles(run) {
97
+ const listing = run(['ls-files']);
98
+ if (listing === '') return null;
99
+ return new Set(listing.split('\n').filter((line) => line !== ''));
100
+ }
101
+
102
+ /**
103
+ * Which package manager this repo actually uses.
104
+ *
105
+ * A lockfile being ON DISK is not evidence that it is the repo's lockfile.
106
+ * Measured on a real install: the repo had both `pnpm-lock.yaml` and
107
+ * `package-lock.json`, `pnpm-lock.yaml` was gitignored and untracked, and
108
+ * `.gitignore` said in words that the npm lockfile was authoritative. Picking
109
+ * by disk order chose pnpm, and then EVERY validation command was wrong —
110
+ * `pnpm lint`, `pnpm typecheck`, `pnpm test` in a repo with no pnpm.
111
+ *
112
+ * So git is asked which lockfiles it tracks, and a tracked one wins. The
113
+ * distinction that keeps this honest: `git ls-files` returning NOTHING means
114
+ * git could not answer (no repository, nothing committed yet), and no
115
+ * conclusion is drawn from it. A non-empty listing that omits a lockfile is
116
+ * positive evidence that the file is ignored or untracked, and that is the
117
+ * only case that changes the answer.
118
+ */
119
+ export function detectPackageManager(dir, run = gitRunner(dir)) {
120
+ const present = Object.entries(LOCKFILES).filter(([file]) => existsSync(join(dir, file)));
121
+ if (present.length > 0) {
122
+ const tracked = trackedFiles(run);
123
+ if (tracked === null) {
124
+ const [file, manager] = present[0];
125
+ return provenance(manager, `lockfile: ${file}`, 0.95);
126
+ }
127
+ const inGit = present.filter(([file]) => tracked.has(file));
128
+ if (inGit.length > 0) {
129
+ const [file, manager] = inGit[0];
130
+ const ignored = present.filter(([f]) => !tracked.has(f)).map(([f]) => f);
131
+ const source =
132
+ ignored.length === 0
133
+ ? `lockfile: ${file}`
134
+ : `lockfile: ${file} — tracked by git, unlike ${ignored.join(', ')}, which git does not track`;
135
+ return provenance(manager, source, 0.95);
136
+ }
137
+ // On disk, none of them committed. Something is the answer here, but the
138
+ // repository is not saying which — so it is a flagged guess, not a fact.
139
+ const [file, manager] = present[0];
140
+ return provenance(
141
+ manager,
142
+ `lockfile: ${file}, but git tracks no lockfile at all — this one may be ignored`,
143
+ 0.4,
144
+ true,
145
+ );
96
146
  }
97
147
  if (existsSync(join(dir, 'package.json'))) {
98
148
  return provenance('npm', 'package.json with no lockfile — npm assumed', 0.5, true);
@@ -117,12 +167,77 @@ export function readPackageJson(dir) {
117
167
  * gate is noise. When nothing can be read, the list comes back empty and
118
168
  * flagged rather than filled in with `npm test`.
119
169
  */
170
+ /**
171
+ * Suffixes a repo uses for "the same thing, but it exits". Tried in order.
172
+ */
173
+ export const RUN_ONCE_SUFFIXES = Object.freeze([':run', ':ci', ':once']);
174
+
175
+ /**
176
+ * Why this script would never return, or null.
177
+ *
178
+ * This is the single most dangerous inference the scanner can make. A
179
+ * validation command that watches instead of exiting does not fail the agent
180
+ * that runs it — it HANGS it, with no output and no timeout, and the operator
181
+ * sees a session that has simply stopped. Measured on a real install: setup
182
+ * wrote `pnpm test`, whose script was bare `vitest`, and every agent handed
183
+ * that gate would have waited forever.
184
+ *
185
+ * Deliberately narrow, because a false positive here drops a real test command
186
+ * out of the config. `jest` is NOT listed: it runs once by default and only
187
+ * watches when told to, so the explicit-flag rule below already covers it.
188
+ */
189
+ export function watchModeProblem(script) {
190
+ const text = String(script);
191
+ if (/(^|[\s;&|])nodemon(\s|$)/.test(text)) return 'runs `nodemon`, which watches and never exits';
192
+ if (/--watch(All|-all)?\b(?!=false)/.test(text)) return 'passes --watch, so it never exits';
193
+ // `vitest` with no subcommand is watch mode outside CI. `vitest run`,
194
+ // `vitest bench` and an explicit `--run` all terminate.
195
+ if (/(^|[\s;&|])vitest(\s|$)/.test(text) && !/(^|\s)vitest\s+(run|bench|related)\b/.test(text) && !/--run\b/.test(text)) {
196
+ return 'is bare `vitest`, which defaults to watch mode and never exits';
197
+ }
198
+ return null;
199
+ }
200
+
201
+ /**
202
+ * The script to actually run for `name`, avoiding a watcher.
203
+ *
204
+ * Returns `{ script, note }` — `script` null when every candidate watches,
205
+ * which is a command deliberately LEFT OUT rather than written into a config
206
+ * where it would hang the first agent that trusted it.
207
+ */
208
+ export function resolveValidationScript(scripts, name) {
209
+ const problem = watchModeProblem(scripts[name]);
210
+ if (problem === null) return { script: name, note: null };
211
+ for (const suffix of RUN_ONCE_SUFFIXES) {
212
+ const alternate = `${name}${suffix}`;
213
+ if (typeof scripts[alternate] === 'string' && watchModeProblem(scripts[alternate]) === null) {
214
+ return { script: alternate, note: `\`${name}\` ${problem}, so \`${alternate}\` is used instead` };
215
+ }
216
+ }
217
+ return { script: null, note: `\`${name}\` ${problem}, and no run-once variant exists — left out` };
218
+ }
219
+
120
220
  export function detectValidation(dir, pkg, manager) {
121
221
  const scripts = pkg?.scripts;
122
222
  if (scripts && typeof scripts === 'object') {
123
223
  const prefix = RUN_PREFIX[manager] ?? 'npm run';
124
- const found = VALIDATION_SCRIPTS.filter((name) => typeof scripts[name] === 'string').map((n) => `${prefix} ${n}`);
125
- if (found.length > 0) return provenance(found, `package.json scripts: ${found.length} of the usual names`, 0.9);
224
+ const found = [];
225
+ const notes = [];
226
+ let dropped = false;
227
+ for (const name of VALIDATION_SCRIPTS) {
228
+ if (typeof scripts[name] !== 'string') continue;
229
+ const { script, note } = resolveValidationScript(scripts, name);
230
+ if (note !== null) notes.push(note);
231
+ if (script === null) dropped = true;
232
+ else found.push(`${prefix} ${script}`);
233
+ }
234
+ if (found.length > 0 || notes.length > 0) {
235
+ const base = `package.json scripts: ${found.length} of the usual names`;
236
+ const source = notes.length === 0 ? base : `${base}. ${notes.join('; ')}`;
237
+ // A dropped command means this repo's validation is INCOMPLETE, and the
238
+ // operator is the only one who can say what to run instead.
239
+ return provenance(found, source, dropped ? 0.6 : 0.9, dropped);
240
+ }
126
241
  }
127
242
  if (existsSync(join(dir, 'Makefile'))) {
128
243
  try {
@@ -200,7 +315,7 @@ export function detectAutonomy(dir, run = gitRunner(dir)) {
200
315
  /** The whole scan. Returns `{config, languages, questions}`. */
201
316
  export function scanRepo(dir, { run = gitRunner(dir) } = {}) {
202
317
  const pkg = readPackageJson(dir);
203
- const manager = detectPackageManager(dir);
318
+ const manager = detectPackageManager(dir, run);
204
319
  const validation = detectValidation(dir, pkg, manager?.value);
205
320
  const autonomy = detectAutonomy(dir, run);
206
321
  const languages = detectLanguages(dir, run);
@@ -235,6 +350,16 @@ export function renderConfig(config) {
235
350
  '# Written by /tyran:setup. Inferred values carry provenance so you can\n' +
236
351
  '# audit where each one came from. Edit freely — your edits win.\n' +
237
352
  '#\n' +
353
+ '# EDITING BY HAND: Tyran parses a small YAML subset. Block scalars (>- and\n' +
354
+ '# |-) are REJECTED, and a file this parser cannot read makes the policy\n' +
355
+ '# gate refuse every write in this repository. Keep each `source:` on ONE\n' +
356
+ '# line, single-quoted, however long it gets.\n' +
357
+ '#\n' +
358
+ '# COMMIT THIS DIRECTORY. `.tyran/` is your repo\'s data, and an untracked\n' +
359
+ '# one does not reach `git worktree add` — agents then run there with no\n' +
360
+ '# autonomy class at all, which is the boundary silently missing exactly\n' +
361
+ '# where the most agents run.\n' +
362
+ '#\n' +
238
363
  '# Validate: node scripts/schema.mjs validate config .tyran/config.yaml\n' +
239
364
  '# Resolve a role to a model: node scripts/tiers.mjs --role reviewer\n' +
240
365
  '\n' +
@@ -242,6 +367,140 @@ export function renderConfig(config) {
242
367
  );
243
368
  }
244
369
 
370
+ // ------------------------------------------------------- bootstrapping the policy
371
+
372
+ /**
373
+ * The directory whose existence ARMS the policy gate, and the file that gate
374
+ * needs in order to answer anything.
375
+ */
376
+ export const TYRAN_DIR = '.tyran';
377
+ export const POLICY_PATH = '.tyran/policies/autonomy.yaml';
378
+
379
+ /**
380
+ * Seeding the policy is part of writing the config, and it is not optional.
381
+ *
382
+ * The policy gate is silent in a repository with no `.tyran/` directory — no
383
+ * adoption, nothing to say — and REFUSES every write in a repository that has
384
+ * one but no `.tyran/policies/autonomy.yaml`, because a boundary it cannot
385
+ * read must never read as an open one (ADR-22). Both halves are right. Their
386
+ * seam was not: `--write .tyran/config.yaml` created the directory and nothing
387
+ * created the policy, so setup's own first command moved the repo from
388
+ * "silent" to "refuses everything" — including the write that would have
389
+ * installed the missing file. Measured on a real install: the setup session
390
+ * ended with the operator being handed a `mkdir` and a `cp` to run by hand.
391
+ *
392
+ * Seeding it here is bootstrap, not self-authorization, and the difference is
393
+ * mechanical rather than a matter of intent:
394
+ *
395
+ * - it only ever CREATES. An existing policy is never read, never merged,
396
+ * never overwritten — `existsSync` returns first. So this cannot weaken a
397
+ * boundary that exists, which is the whole content of KERNEL;
398
+ * - what it writes is the shipped template byte for byte, which is the
399
+ * strictest default Tyran has (`default: GATED`, `hooks/**` and
400
+ * `.tyran/policies/**` KERNEL). There is no arrangement of inputs that
401
+ * makes it write something more permissive;
402
+ * - it is validated before it lands, so a damaged template becomes a loud
403
+ * exit 2 rather than a repository nobody can write to.
404
+ *
405
+ * Editing the policy afterwards is still human-only, by hand, and the gate
406
+ * still refuses `Write`, `Edit` and any shell command that names the path.
407
+ */
408
+ export class BootstrapError extends Error {
409
+ constructor(message, remedy) {
410
+ super(message);
411
+ this.name = 'BootstrapError';
412
+ this.remedy = remedy;
413
+ }
414
+ }
415
+
416
+ /** The shipped template, from this script's REAL location (see isMainModule). */
417
+ export function policyTemplatePath() {
418
+ const self = canonicalPath(fileURLToPath(import.meta.url));
419
+ return join(dirname(self), '..', 'templates', 'policies', 'autonomy.yaml');
420
+ }
421
+
422
+ /** True when writing `path` would create or populate the repo's `.tyran/`. */
423
+ export function underTyranDir(path, repoRoot) {
424
+ const rel = relative(resolve(repoRoot), resolve(path)).replace(/\\/g, '/');
425
+ return rel === TYRAN_DIR || rel.startsWith(`${TYRAN_DIR}/`);
426
+ }
427
+
428
+ /** The ancestors of `leaf` up to and including `stopAt` that do not exist yet. */
429
+ function absentAncestors(leaf, stopAt) {
430
+ const absent = [];
431
+ let cursor = resolve(leaf);
432
+ const root = resolve(stopAt);
433
+ while (cursor !== root && cursor !== dirname(cursor)) {
434
+ if (!existsSync(cursor)) absent.push(cursor);
435
+ cursor = dirname(cursor);
436
+ }
437
+ return absent;
438
+ }
439
+
440
+ /**
441
+ * Install `.tyran/policies/autonomy.yaml` from the shipped template if it is
442
+ * absent. Returns `{ path, status: 'created' | 'present' }`.
443
+ *
444
+ * On failure it removes the directories it created before throwing. That
445
+ * cleanup is the point rather than tidiness: a half-finished bootstrap leaves
446
+ * `.tyran/` on disk with no policy under it, which is exactly the state that
447
+ * refuses every subsequent write — the bug this function exists to prevent,
448
+ * reintroduced by its own error path.
449
+ */
450
+ export function ensureAutonomyPolicy(repoRoot, { templatePath = policyTemplatePath() } = {}) {
451
+ const root = resolve(repoRoot);
452
+ const path = join(root, ...POLICY_PATH.split('/'));
453
+ if (existsSync(path)) return { path, status: 'present' };
454
+
455
+ let text;
456
+ try {
457
+ text = readFileSync(templatePath, 'utf8');
458
+ } catch (error) {
459
+ throw new BootstrapError(
460
+ `the shipped policy template is unreadable at ${escapeInvisible(templatePath)}: ${error.message}`,
461
+ 'reinstall the plugin — this file ships with it, and without it a repository cannot be set up at all',
462
+ );
463
+ }
464
+
465
+ // Validated BEFORE it lands. A template that fails its own schema installs a
466
+ // repository the gate refuses on every write, and the operator then debugs a
467
+ // policy they never wrote.
468
+ let errors;
469
+ try {
470
+ errors = validatePolicy(parse(text));
471
+ } catch (error) {
472
+ errors = [`the template is not parseable YAML (${error.name})`];
473
+ }
474
+ if (errors.length > 0) {
475
+ throw new BootstrapError(
476
+ `the shipped policy template does not validate (${errors.length} finding(s); the first is ` +
477
+ `${escapeInvisible(String(errors[0]))})`,
478
+ 'this is a defect in the plugin, not in your repository. Report it rather than hand-editing the template.',
479
+ );
480
+ }
481
+
482
+ const created = absentAncestors(dirname(path), root);
483
+ try {
484
+ mkdirSync(dirname(path), { recursive: true });
485
+ writeFileSync(path, text);
486
+ } catch (error) {
487
+ for (const dir of created) {
488
+ // Deepest first, and only while empty — rmdirSync refuses a directory
489
+ // with anything in it, which is the behaviour to want here.
490
+ try {
491
+ rmdirSync(dir);
492
+ } catch {
493
+ break;
494
+ }
495
+ }
496
+ throw new BootstrapError(
497
+ `could not write ${escapeInvisible(path)}: ${error.message}`,
498
+ 'check the directory permissions; nothing was left behind, so re-running this command is safe',
499
+ );
500
+ }
501
+ return { path, status: 'created' };
502
+ }
503
+
245
504
  function main() {
246
505
  const args = process.argv.slice(2);
247
506
  const flag = (name, fallback) => {
@@ -254,10 +513,34 @@ function main() {
254
513
  process.exit(2);
255
514
  }
256
515
 
516
+ // Repair mode: seed the policy and do nothing else. This is what `doctor`
517
+ // hands an operator whose `.tyran/` predates this bootstrap, and it names no
518
+ // protected path on its command line, so it is a remedy an agent can run
519
+ // rather than a chore an agent has to delegate back to a human.
520
+ if (args.includes('--ensure-policy')) {
521
+ const seeded = seedPolicy(dir);
522
+ console.error(
523
+ seeded.status === 'created'
524
+ ? `scan-repo: created ${escapeInvisible(seeded.path)} from the shipped template`
525
+ : `scan-repo: ${escapeInvisible(seeded.path)} already exists — left untouched`,
526
+ );
527
+ return;
528
+ }
529
+
257
530
  const result = scanRepo(dir);
258
531
  const target = flag('write', null);
259
532
  if (target !== null) {
260
533
  const path = resolve(target);
534
+ // The boundary goes in FIRST. Between the mkdir that creates `.tyran/` and
535
+ // the write that puts a policy under it, every tool call in the session is
536
+ // refused; ordering it this way makes that window empty instead of
537
+ // permanent.
538
+ if (underTyranDir(path, dir)) {
539
+ const seeded = seedPolicy(dir);
540
+ if (seeded.status === 'created') {
541
+ console.error(`scan-repo: created ${escapeInvisible(seeded.path)} from the shipped template`);
542
+ }
543
+ }
261
544
  try {
262
545
  mkdirSync(dirname(path), { recursive: true });
263
546
  writeFileSync(path, renderConfig(result.config));
@@ -270,6 +553,18 @@ function main() {
270
553
  console.log(JSON.stringify(result, null, 2));
271
554
  }
272
555
 
556
+ /** `ensureAutonomyPolicy`, with a BootstrapError turned into exit 2. */
557
+ function seedPolicy(dir) {
558
+ try {
559
+ return ensureAutonomyPolicy(dir);
560
+ } catch (error) {
561
+ if (!(error instanceof BootstrapError)) throw error;
562
+ console.error(`scan-repo: ${error.message}`);
563
+ console.error(`scan-repo: ${error.remedy}`);
564
+ process.exit(2);
565
+ }
566
+ }
567
+
273
568
  function canonicalPath(path) {
274
569
  const abs = resolve(path);
275
570
  try {
@@ -133,7 +133,28 @@ function stripComment(line, lineNo) {
133
133
  // YAML loader (`a: don't ask # why`) — refuse rather than guess
134
134
  // (review E2S2 note 1).
135
135
  if (inSingle || inDouble) {
136
- throw new YamlLiteError('unbalanced quote — quote the whole value if it contains an apostrophe', lineNo);
136
+ // Name the construct, not one guess at its cause.
137
+ //
138
+ // This said "quote the whole value if it contains an apostrophe" for every
139
+ // unbalanced quote there is. Measured on a real install: an operator wrote
140
+ // a long `source:` as a multi-line double-quoted string, got told to worry
141
+ // about an apostrophe that was not there, and burned three round-trips
142
+ // looking for one. A diagnosis that confidently names the wrong cause is
143
+ // worse than one that names none — it sends the reader somewhere else.
144
+ //
145
+ // The two cases are distinguishable from what is already known here: a
146
+ // value that OPENS with the quote still in flight was never closed on this
147
+ // line, which in a subset with no multi-line scalars is the whole answer.
148
+ const opener = inSingle ? "'" : '"';
149
+ const body = line.slice(line.indexOf(':') + 1).trim();
150
+ const opensWithQuote = body.startsWith(opener);
151
+ throw new YamlLiteError(
152
+ opensWithQuote
153
+ ? `a ${opener}-quoted value that never closes on this line. This subset has no multi-line ` +
154
+ 'scalars (no >-, no |), so put the whole value on one line however long it gets'
155
+ : `unbalanced ${opener} — if the value contains an apostrophe, quote the whole value`,
156
+ lineNo,
157
+ );
137
158
  }
138
159
  return cut === -1 ? line : line.slice(0, cut);
139
160
  }
@@ -236,6 +236,16 @@ open one.
236
236
  sets only; a git worktree per agent; shared zones (per the manifest in
237
237
  `PLAN.md`) are APPEND-ONLY and serialized; a branch per story; rebase
238
238
  before merge.
239
+ - **A worktree without `.tyran/` is an UNGATED worktree.** `git worktree
240
+ add` carries tracked files only, so an uncommitted `.tyran/` does not
241
+ reach one — and the policy gate is deliberately silent in a repo that has
242
+ no `.tyran/` at all. The agents there then run with no autonomy class and
243
+ no path classes: nothing fails, the boundary is simply absent, in the one
244
+ place you are running the most agents. Measured on a real install: four
245
+ worktrees, four ungated implementers. So **check every worktree you
246
+ create** and, if `.tyran/` is missing, get it committed on the main
247
+ checkout first rather than copying it in — a copy makes four divergent
248
+ configs, which is the same defect one layer down.
239
249
  - **Leases are files.** Every worktree and every heavy slot has
240
250
  `.tyran/initiatives/<slug>/locks/<resource>.lease` holding holder,
241
251
  purpose, story and expiry. A handoff BEGINS by taking the lease — an
@@ -20,6 +20,27 @@ an autonomy class inferred from how the repository is really worked. Every
20
20
  value carries the fact that produced it and a confidence, so months later
21
21
  "why does this repo think it is P2" has an answer in the file itself.
22
22
 
23
+ **It writes two files, and the second one is why this command is not
24
+ optional.** Before the config, it installs `.tyran/policies/autonomy.yaml`
25
+ from the shipped template. The policy gate is silent in a repository with no
26
+ `.tyran/` directory and refuses every write in a repository that has one
27
+ without a policy under it — so creating the config alone moves the repo from
28
+ "unmanaged" to "refuses everything", including the write that would have
29
+ fixed it. Setup did exactly that once, and the session ended with a `mkdir`
30
+ and a `cp` handed to the operator.
31
+
32
+ Installing it is bootstrap, not an agent granting itself permission: it only
33
+ ever creates, never overwrites, and what it writes is the strictest template
34
+ Tyran ships. Editing that file afterwards is human-only and stays so — the
35
+ gate refuses `Write`, `Edit`, and any shell command that names the path.
36
+
37
+ If you find a repository already in the broken state — `.tyran/` present, no
38
+ policy — this repairs it without touching anything else:
39
+
40
+ ```bash
41
+ node "${CLAUDE_PLUGIN_ROOT}/scripts/scan-repo.mjs" --dir . --ensure-policy
42
+ ```
43
+
23
44
  **It never infers `P3`.** No arrangement of files is evidence that a person
24
45
  meant to let an agent merge to main and deploy to production. That is a
25
46
  decision someone makes in words, and if the operator wants it they will say
@@ -71,9 +92,41 @@ someone's working tree unannounced is a bad way to meet a tool.
71
92
  node "${CLAUDE_PLUGIN_ROOT}/scripts/schema.mjs" validate config .tyran/config.yaml
72
93
  node "${CLAUDE_PLUGIN_ROOT}/scripts/tiers.mjs" --config .tyran/config.yaml
73
94
  node "${CLAUDE_PLUGIN_ROOT}/scripts/doctor.mjs" --hooks
95
+ node "${CLAUDE_PLUGIN_ROOT}/scripts/doctor.mjs" --state --dir .tyran
96
+ ```
97
+
98
+ The `--state` run is the one that answers "is this repo actually usable now",
99
+ and it is also how you validate the policy. Do **not** reach for
100
+ `schema.mjs validate policy .tyran/policies/autonomy.yaml`: the gate refuses
101
+ any shell command that names a path under `.tyran/policies/**`, including
102
+ that one. `doctor --state` validates the same file and names only `.tyran`.
103
+ A `policy-missing` finding there is an **error**, not a note — it means every
104
+ write in the repo is being refused.
105
+
106
+ Report in six lines: what was detected, what was inferred and from what, what
107
+ you asked about, that both `.tyran/` files are in place and validate, whether
108
+ the shortcut was installed, and what to run next. Paste the raw output of the
109
+ commands above — the evidence contract binds you exactly as it binds every
110
+ agent you will spawn.
111
+
112
+ ## 6. Tell the operator to commit `.tyran/`, and say why
113
+
114
+ End by asking for one commit of `.tyran/`. This is not tidiness, and the
115
+ reason is the one failure mode of this whole setup that is **silent**:
116
+
117
+ ```bash
118
+ git add .tyran && git commit -m "chore: adopt Tyran"
74
119
  ```
75
120
 
76
- Report in five lines: what was detected, what was inferred and from what,
77
- what you asked about, whether the shortcut was installed, and what to run
78
- next. Paste the raw output of the three commands above — the evidence
79
- contract binds you exactly as it binds every agent you will spawn.
121
+ `git worktree add` does not carry untracked files. Tyran's parallel model is
122
+ worktrees, so an uncommitted `.tyran/` means every worktree the conductor
123
+ creates has no config and no policy — and a repo with no `.tyran/` is a repo
124
+ the policy gate deliberately says nothing about. The agents running in those
125
+ worktrees then have no autonomy class and no path classes at all. Nothing
126
+ fails; the boundary is simply not there, in exactly the place the most agents
127
+ run. Measured on a real install: four worktrees, four ungated implementers.
128
+
129
+ Until it is committed there is a second hazard worth naming in your report: an
130
+ agent that runs `git add -A` on a story branch sweeps the untracked `.tyran/`
131
+ into that branch, and a four-way parallel run ends with four conflicting
132
+ copies of the config. Stage explicit paths until this commit exists.
@@ -32,9 +32,21 @@ rules:
32
32
  class: AUTO
33
33
  reason: the execution journal is append-only and written by the conductor itself
34
34
 
35
+ # Was GATED. The cost was measured on a real install: setup had inferred
36
+ # `pnpm test`, which in that repo is bare `vitest` — it never exits, so every
37
+ # agent would have hung forever. The session that FOUND that could not fix it,
38
+ # and handed the operator a heredoc to paste. A boundary that blocks the repair
39
+ # of a footgun it created is one the user edits away, and then it protects
40
+ # nothing at all.
41
+ #
42
+ # What this gives up, in one sentence, because an unstated cost is a false
43
+ # guarantee: `autonomy:` lives in this file too, so an agent can now raise its
44
+ # own deployment class. Set this rule back to GATED if that trade is wrong for
45
+ # your repo — and read `docs/policy-gate.md`, which measured the same
46
+ # escalation happening under GATED anyway wherever `Write` is allow-listed.
35
47
  - path: .tyran/config.yaml
36
- class: GATED
37
- reason: autonomy class and model routing live here, so this file sets both what Tyran may do and what it may spend
48
+ class: AUTO
49
+ reason: mostly a description of this repo — validation commands, package manager, shared zones — and an agent that finds it wrong is what is best placed to correct it
38
50
 
39
51
  - path: .claude/agents/**
40
52
  class: GATED