2a0b1f7 · 8703d33 · 4bbe55d · 454b745 · 3b6e2e0 · 96ff4b1
@@ -81,6 +81,7 @@
for a tunnel and a forge that sleeps.
| `hooks` | `{ repo, action, url, secret, kinds, id }`: `list`, `add` (`secret`, 16–200 characters, is required) or `remove` the repository's webhooks, as `…/hooks`; an admin-scope token on a repository you administer, and a secret is never shown | yes |
| `invites` | `{ action, id? }`: `list` the outstanding invites (state and expiry, never a link) or `revoke` one by `id`; a person admin with an admin token, stricter than the API's list, which an `ro` token may read | yes |
| `issue` | `{ repo, number }`: one issue's title, state, closeReason, body, labels, assignee, comments (author, authorIsAgent, body, at) and blockedBy/blocks, as `GET …/issues/:number`; a sealed one you cannot see is not found |  |
| `issue_teams` | `{ repo, number, teams? }`: the crew teams an issue names; with `teams` (at most two slugs, `[]` clears) names them, and each reviews the closing pull | yes |
| `labels` | `{ repo, number, labels }`: set an issue's or a pull's labels to exactly these (`[]` clears); only labels the repository has, by a writer | yes |
| `limb` | `{ app, issue, action, reason?, pillar?, operation?, dryRun? }`: defer, resume or close a whole limb as not doing, or undo an operation; plans unless `dryRun` is false, and applying takes an admin token | yes |
| `limits` | `{ key?, value?, dryRun? }` (forge admin): every limit the forge's owner can hit, with its value, default and bounds; with `key` and `value` (`""` restores the default), planned first (`dryRun` is true unless `false`); a value out of bounds is refused, applying needs an admin-scope token | yes |
      
        sentinelcrew
        19d ago
        outdated
      
      Nit: mayBoards refuses anyone who isn't a forge admin, even with dryRun true. The row only mentions the admin token. "a forge admin's; applying takes an admin token" would match subscription_budget. The same applies to model_grant on the next line.
    
@@ -90,6 +91,8 @@
for a tunnel and a forge that sleeps.
| `milestones` | `{ app, issue? }`: how far each thing asked for is (and, with `issue`, whether you watch it: `watching`), from the dependency tree: the plan's limbs with their weights, or one milestone opened (who is building, what is blocked and on what, what merged in 7 days, an estimate, and for an owner `spend`: dollars, changes landed ever, the costliest three issues) |  |
| `mint_invite` | `{ days? }`: mint a single-use invite link that expires in `days` (1–30, default 7) and is shown once; a person admin's, with an admin token | yes |
| `mirror_set` | `{ action, endpoint?, bucket?, region?, accessKey?, secretKey? }` (forge admin, admin-scope token): `set` seals the off-box snapshot mirror, `forget` clears it; the Backups page's rules and refusals, a blank field keeps what was sealed; answers `{ place, says }`, never a key | yes |
| `model_ceiling` | `{ username, dailyUsd, perRunUsd, dryRun? }` (a forge admin who is a person): cap an account's daily and per-run model spend in dollars (both `null` clears); planned unless `dryRun` is false, applying takes an admin token | yes |
      
        sentinelcrew
        18d ago
        outdated
      
      model_ceiling and model_grant both check mayBoards: a forge admin who is a person, never an agent. The row only says 'applying takes an admin token', and even the plan is refused for anyone else. Name the caller.
    
| `model_grant` | `{ username, usdPerMonth, dryRun? }` (a forge admin who is a person): give an account a capped monthly model grant in dollars (0 takes it back); planned unless `dryRun` is false, applying takes an admin token | yes |
| `new_app` | `{ name, description?, owner?, visibility?, agent?, pick? }`: make an app from a sentence; answers `{ repo, url, issue, agent, run }` (`run` is the first run's state, or `none`) and draws the `ui://forge/app` card: name, production URL, first issue, agent, first run; refused as `POST /api/v1/apps` is (unverified, app_limit, no_such_agent, name_taken) | yes |
| `next_steps` | on a spawned forge, your first three steps and whether each is done; empty elsewhere or for anyone but its owner |  |
| `now` | what the crew is doing right now: who, on what, since when, at what cost; a planner that declined the direction and waits for you comes first (`run: null`, its words in `says`) |  |
@@ -120,6 +123,7 @@
for a tunnel and a forge that sleeps.
| `run_log` | `{ repo, run, tail? }`: a run's log, or its last `tail` lines (1–2000), with the REST route's access: a hidden log or unreadable repository is not found |  |
| `runs` | `{ repo, limit? }`: the repository's recent runs, newest first (id, kind, state, issue or pull number, started, ended), 1–50, default 20; read as the runs page reads them |  |
| `say` | speak on an issue or a pull; a working builder hears it as its next step | yes |
      
        sentinelcrew
        18d ago
        outdated
      
      scorecard is adminOnly. Say (forge admin) here, as system and capacity do, so a reader doesn't expect it to answer them.
    
| `scorecard` | (forge admin) no input: the crew's scorecard per agent, all time and 30 days: landed, review rounds, red rate, reverts |  |
| `seal` | `{ repo, number, sealed }`: seal a security report (and the pull that fixes it) so only writers and the crew see it, or unseal it for every reader; a writer's, recorded; with `action: "private"` instead of `sealed`, the private repository a writer fixes a sealed report in, made on first need, with its name and clone URL (the API's `POST …/issues/:number/private`, same rules) | yes |
| `search` | `{ q, kind?, repo? }`: issues, pulls, repositories and code you can read, the same results as /search; `repo` keeps to one |  |
| `secrets` | `{ scope, ref?, action, name?, value? }` (forge admin): `list` the names and timestamps of the forge's sealed secrets, `set` or `remove` one; `scope` is forge, org or repo, `ref` the org or `owner/name`; a value is never read back | yes |
@@ -134,6 +138,7 @@
for a tunnel and a forge that sleeps.
| `subscription_wake` | `{ id }`: use a resting Claude subscription again now, as the Subscriptions page's Wake button does; answers `{ id, says }`; a forge admin, an admin token | yes |
| `system` | the machine the forge runs on, now: CPU, load, memory, disk, network, processes; a forge admin, or an agent granted `telemetry:read` |  |
| `take_updates` | a fork takes its parent's updates when git can do it alone; a conflict is refused. On a spawned forge's code repository it opens the one `forge/take-updates` pull and never moves the default branch | yes |
| `teams` | `{ slug?, charter?, dial?, members?, dryRun? }`: the crew teams (person admins): without `slug` lists them; with it, changes charter, dial (0–100) or members, planned unless `dryRun` is false | yes |
| `telemetry_grant` | `{ agent, on }`: grant an agent `telemetry:read`, or take it back with `false`, as the Telemetry page does; a person admin's, with an admin token; answers `says` | yes |
| `telemetry_history` | `{ hours? }` (1–72, default 6): the node's and the forge's CPU cores, memory and load over the window, `{ hours, points: [{ target, at, cpuCores, cpuCoresMax (the bucket's peak), memoryBytes, load }] }`, at most 300 points a series (15-s samples to 24 h, 5-min beyond, averaged down); a forge admin or `telemetry:read`, as `system`; the machine records no disk history |  |
| `tell` | a sentence becomes an issue, a builder is picked, a run starts on the crew's next tick | yes |
@@ -16,6 +16,7 @@
export const tool: Tool = {
    required: ["repo", "number"],
  },
  writes: true,
  doc: "`{ repo, number, teams? }`: the crew teams an issue names; with `teams` (at most two slugs, `[]` clears) names them, and each reviews the closing pull",
  run(ctx, user, args) {
    const repo = repoArg(ctx, user, args);
    if (isAnswer(repo)) return repo;
@@ -18,6 +18,7 @@
export const tool: Tool = {
    required: ["username", "dailyUsd", "perRunUsd"],
  },
  writes: true,
  doc: "`{ username, dailyUsd, perRunUsd, dryRun? }` (a forge admin who is a person): cap an account's daily and per-run model spend in dollars (both `null` clears); planned unless `dryRun` is false, applying takes an admin token",
  dryByDefault: true,
  run(ctx, user, args) {
    const { username, dailyUsd, perRunUsd } = args;
@@ -17,6 +17,7 @@
export const tool: Tool = {
    required: ["username", "usdPerMonth"],
  },
  writes: true,
  doc: "`{ username, usdPerMonth, dryRun? }` (a forge admin who is a person): give an account a capped monthly model grant in dollars (0 takes it back); planned unless `dryRun` is false, applying takes an admin token",
  dryByDefault: true,
  run(ctx, user, args) {
    const usd = args["usdPerMonth"];
@@ -8,6 +8,7 @@
export const tool: Tool = {
    "Crew scorecard (forge admin): per agent, all time and 30 days: landed, rounds, red rate.",
  inputSchema: { type: "object", properties: {} },
  writes: false,
  doc: "(forge admin) no input: the crew's scorecard per agent, all time and 30 days: landed, review rounds, red rate, reverts",
      
        sentinelcrew
        19d ago
        outdated
      
      Nit: this tool is adminOnly and anyone else gets forbidden, but the row doesn't say so. Suggest: "no input (forge admin): the crew's scorecard per agent…", the way the system row reads.
    
  adminOnly: true,
  run(ctx, user) {
    const card = readScorecards(ctx.app, user);
@@ -17,6 +17,7 @@
export const tool: Tool = {
    },
  },
  writes: true,
  doc: "`{ slug?, charter?, dial?, members?, dryRun? }`: the crew teams (person admins): without `slug` lists them; with it, changes charter, dial (0–100) or members, planned unless `dryRun` is false",
  dryByDefault: true,
  run(ctx, user, args) {
    const read = readTeams(ctx.app, user);
@@ -23,12 +23,7 @@
const undocumented = new Set<string>([
  "asks_decide",
  "asks_proposal",
  "asks_undecide",
  "issue_teams",
  "model_grant",
  "model_ceiling",
  "spend",
  "scorecard",
  "teams",
  "report",
]);
		@@ -81,6 +81,7 @@ for a tunnel and a forge that sleeps.
81	81	\| `hooks` \| `{ repo, action, url, secret, kinds, id }`: `list`, `add` (`secret`, 16–200 characters, is required) or `remove` the repository's webhooks, as `…/hooks`; an admin-scope token on a repository you administer, and a secret is never shown \| yes \|
82	82	\| `invites` \| `{ action, id? }`: `list` the outstanding invites (state and expiry, never a link) or `revoke` one by `id`; a person admin with an admin token, stricter than the API's list, which an `ro` token may read \| yes \|
83	83	\| `issue` \| `{ repo, number }`: one issue's title, state, closeReason, body, labels, assignee, comments (author, authorIsAgent, body, at) and blockedBy/blocks, as `GET …/issues/:number`; a sealed one you cannot see is not found \| \|
	84	\| `issue_teams` \| `{ repo, number, teams? }`: the crew teams an issue names; with `teams` (at most two slugs, `[]` clears) names them, and each reviews the closing pull \| yes \|
84	85	\| `labels` \| `{ repo, number, labels }`: set an issue's or a pull's labels to exactly these (`[]` clears); only labels the repository has, by a writer \| yes \|
85	86	\| `limb` \| `{ app, issue, action, reason?, pillar?, operation?, dryRun? }`: defer, resume or close a whole limb as not doing, or undo an operation; plans unless `dryRun` is false, and applying takes an admin token \| yes \|
86	87	\| `limits` \| `{ key?, value?, dryRun? }` (forge admin): every limit the forge's owner can hit, with its value, default and bounds; with `key` and `value` (`""` restores the default), planned first (`dryRun` is true unless `false`); a value out of bounds is refused, applying needs an admin-scope token \| yes \|
		sentinelcrew 19d ago outdated Nit: `mayBoards` refuses anyone who isn't a forge admin, even with dryRun true. The row only mentions the admin token. "a forge admin's; applying takes an admin token" would match `subscription_budget`. The same applies to `model_grant` on the next line.
		@@ -90,6 +91,8 @@ for a tunnel and a forge that sleeps.
90	91	\| `milestones` \| `{ app, issue? }`: how far each thing asked for is (and, with `issue`, whether you watch it: `watching`), from the dependency tree: the plan's limbs with their weights, or one milestone opened (who is building, what is blocked and on what, what merged in 7 days, an estimate, and for an owner `spend`: dollars, changes landed ever, the costliest three issues) \| \|
91	92	\| `mint_invite` \| `{ days? }`: mint a single-use invite link that expires in `days` (1–30, default 7) and is shown once; a person admin's, with an admin token \| yes \|
92	93	\| `mirror_set` \| `{ action, endpoint?, bucket?, region?, accessKey?, secretKey? }` (forge admin, admin-scope token): `set` seals the off-box snapshot mirror, `forget` clears it; the Backups page's rules and refusals, a blank field keeps what was sealed; answers `{ place, says }`, never a key \| yes \|
	94	\| `model_ceiling` \| `{ username, dailyUsd, perRunUsd, dryRun? }` (a forge admin who is a person): cap an account's daily and per-run model spend in dollars (both `null` clears); planned unless `dryRun` is false, applying takes an admin token \| yes \|
		sentinelcrew 18d ago outdated model_ceiling and model_grant both check `mayBoards`: a forge admin who is a person, never an agent. The row only says 'applying takes an admin token', and even the plan is refused for anyone else. Name the caller.
	95	\| `model_grant` \| `{ username, usdPerMonth, dryRun? }` (a forge admin who is a person): give an account a capped monthly model grant in dollars (0 takes it back); planned unless `dryRun` is false, applying takes an admin token \| yes \|
93	96	\| `new_app` \| `{ name, description?, owner?, visibility?, agent?, pick? }`: make an app from a sentence; answers `{ repo, url, issue, agent, run }` (`run` is the first run's state, or `none`) and draws the `ui://forge/app` card: name, production URL, first issue, agent, first run; refused as `POST /api/v1/apps` is (unverified, app_limit, no_such_agent, name_taken) \| yes \|
94	97	\| `next_steps` \| on a spawned forge, your first three steps and whether each is done; empty elsewhere or for anyone but its owner \| \|
95	98	\| `now` \| what the crew is doing right now: who, on what, since when, at what cost; a planner that declined the direction and waits for you comes first (`run: null`, its words in `says`) \| \|
		@@ -120,6 +123,7 @@ for a tunnel and a forge that sleeps.
120	123	\| `run_log` \| `{ repo, run, tail? }`: a run's log, or its last `tail` lines (1–2000), with the REST route's access: a hidden log or unreadable repository is not found \| \|
121	124	\| `runs` \| `{ repo, limit? }`: the repository's recent runs, newest first (id, kind, state, issue or pull number, started, ended), 1–50, default 20; read as the runs page reads them \| \|
122	125	\| `say` \| speak on an issue or a pull; a working builder hears it as its next step \| yes \|
		sentinelcrew 18d ago outdated scorecard is `adminOnly`. Say `(forge admin)` here, as `system` and `capacity` do, so a reader doesn't expect it to answer them.
	126	\| `scorecard` \| (forge admin) no input: the crew's scorecard per agent, all time and 30 days: landed, review rounds, red rate, reverts \| \|
123	127	\| `seal` \| `{ repo, number, sealed }`: seal a security report (and the pull that fixes it) so only writers and the crew see it, or unseal it for every reader; a writer's, recorded; with `action: "private"` instead of `sealed`, the private repository a writer fixes a sealed report in, made on first need, with its name and clone URL (the API's `POST …/issues/:number/private`, same rules) \| yes \|
124	128	\| `search` \| `{ q, kind?, repo? }`: issues, pulls, repositories and code you can read, the same results as /search; `repo` keeps to one \| \|
125	129	\| `secrets` \| `{ scope, ref?, action, name?, value? }` (forge admin): `list` the names and timestamps of the forge's sealed secrets, `set` or `remove` one; `scope` is forge, org or repo, `ref` the org or `owner/name`; a value is never read back \| yes \|
		@@ -134,6 +138,7 @@ for a tunnel and a forge that sleeps.
134	138	\| `subscription_wake` \| `{ id }`: use a resting Claude subscription again now, as the Subscriptions page's Wake button does; answers `{ id, says }`; a forge admin, an admin token \| yes \|
135	139	\| `system` \| the machine the forge runs on, now: CPU, load, memory, disk, network, processes; a forge admin, or an agent granted `telemetry:read` \| \|
136	140	\| `take_updates` \| a fork takes its parent's updates when git can do it alone; a conflict is refused. On a spawned forge's code repository it opens the one `forge/take-updates` pull and never moves the default branch \| yes \|
	141	\| `teams` \| `{ slug?, charter?, dial?, members?, dryRun? }`: the crew teams (person admins): without `slug` lists them; with it, changes charter, dial (0–100) or members, planned unless `dryRun` is false \| yes \|
137	142	\| `telemetry_grant` \| `{ agent, on }`: grant an agent `telemetry:read`, or take it back with `false`, as the Telemetry page does; a person admin's, with an admin token; answers `says` \| yes \|
138	143	\| `telemetry_history` \| `{ hours? }` (1–72, default 6): the node's and the forge's CPU cores, memory and load over the window, `{ hours, points: [{ target, at, cpuCores, cpuCoresMax (the bucket's peak), memoryBytes, load }] }`, at most 300 points a series (15-s samples to 24 h, 5-min beyond, averaged down); a forge admin or `telemetry:read`, as `system`; the machine records no disk history \| \|
139	144	\| `tell` \| a sentence becomes an issue, a builder is picked, a run starts on the crew's next tick \| yes \|
		@@ -16,6 +16,7 @@ export const tool: Tool = {
16	16	required: ["repo", "number"],
17	17	},
18	18	writes: true,
	19	doc: "`{ repo, number, teams? }`: the crew teams an issue names; with `teams` (at most two slugs, `[]` clears) names them, and each reviews the closing pull",
19	20	run(ctx, user, args) {
20	21	const repo = repoArg(ctx, user, args);
21	22	if (isAnswer(repo)) return repo;
		@@ -18,6 +18,7 @@ export const tool: Tool = {
18	18	required: ["username", "dailyUsd", "perRunUsd"],
19	19	},
20	20	writes: true,
	21	doc: "`{ username, dailyUsd, perRunUsd, dryRun? }` (a forge admin who is a person): cap an account's daily and per-run model spend in dollars (both `null` clears); planned unless `dryRun` is false, applying takes an admin token",
21	22	dryByDefault: true,
22	23	run(ctx, user, args) {
23	24	const { username, dailyUsd, perRunUsd } = args;
		@@ -17,6 +17,7 @@ export const tool: Tool = {
17	17	required: ["username", "usdPerMonth"],
18	18	},
19	19	writes: true,
	20	doc: "`{ username, usdPerMonth, dryRun? }` (a forge admin who is a person): give an account a capped monthly model grant in dollars (0 takes it back); planned unless `dryRun` is false, applying takes an admin token",
20	21	dryByDefault: true,
21	22	run(ctx, user, args) {
22	23	const usd = args["usdPerMonth"];
		@@ -8,6 +8,7 @@ export const tool: Tool = {
8	8	"Crew scorecard (forge admin): per agent, all time and 30 days: landed, rounds, red rate.",
9	9	inputSchema: { type: "object", properties: {} },
10	10	writes: false,
	11	doc: "(forge admin) no input: the crew's scorecard per agent, all time and 30 days: landed, review rounds, red rate, reverts",
		sentinelcrew 19d ago outdated Nit: this tool is `adminOnly` and anyone else gets `forbidden`, but the row doesn't say so. Suggest: "no input (forge admin): the crew's scorecard per agent…", the way the `system` row reads.
11	12	adminOnly: true,
12	13	run(ctx, user) {
13	14	const card = readScorecards(ctx.app, user);
		@@ -17,6 +17,7 @@ export const tool: Tool = {
17	17	},
18	18	},
19	19	writes: true,
	20	doc: "`{ slug?, charter?, dial?, members?, dryRun? }`: the crew teams (person admins): without `slug` lists them; with it, changes charter, dial (0–100) or members, planned unless `dryRun` is false",
20	21	dryByDefault: true,
21	22	run(ctx, user, args) {
22	23	const read = readTeams(ctx.app, user);
		@@ -23,12 +23,7 @@ const undocumented = new Set<string>([
23	23	"asks_decide",
24	24	"asks_proposal",
25	25	"asks_undecide",
26		"issue_teams",
27		"model_grant",
28		"model_ceiling",
29	26	"spend",
30		"scorecard",
31		"teams",
32	27	"report",
33	28	]);
34	29