surshar / claude-code

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

Built by Anthropic · github.com/anthropics

claude-code / mods / sec-default / README.md

Alice T'Poteatsec-default: the system prompt's sections continue past the user tier (#97241) 684800b · 25 days ago History

188 lines · 11.4 KBPreview CodeRaw

1	# sec-default
2
3	The security default for organizations. Function hooks give every plugin a
4	say on every event, in chain order, and the plugins a person installs sit
5	in the user tier, beneath the organization's prepend tier and above its
6	append tier. Some of what an organization sets today (its classic hooks,
7	its managed CLAUDE.md and rules, its settings, its MCP allowlist, the deny
8	rules in force on its machines) was never within a person's reach before
9	function hooks; seated outermost, this plugin keeps exactly those out of the
10	user tier's reach and adds no policy of its own. Everything else passes
11	through untouched.
12
13	It has three moves and nothing else: continue past the user tier
14	(`next.to(e, "append")`), refuse a user-tier caller or module by name
15	(`{ deny }` when `next.origin.tier` is `user`, `{ refuse }` when a module's
16	pinned `e.tier` is), or pass (`next(e)`). A subject's provenance is the
17	event's pinned `e.provider`; policy is read through
18	`$.settings.read({ source: "policy" })`, one read serving a burst of tool
19	calls; both fail closed, so an unreadable policy counts as a policy in force.
20
21	`hooks/register.ts` is the module; `hooks/policy/` reads the managed
22	settings it decides by.
23
24	## The rows
25
26	\| event \| from the outermost seat \|
27	\| --- \| --- \|
28	\| `classic.*` \| Continue past the user tier: the organization's settings hooks see the engine's input and their answer stands. \|
29	\| `prompt.section`, `prompt.context`, `skill.prompt`, `attribution.text` \| Continue past the user tier: managed CLAUDE.md, rules and policy skills reach the model as written. A person's plugins keep `prompt.submit` and its additive context. \|
30	\| `prompt.compose` \| Continue past the user tier: the system prompt's list of sections is what the organization's tiers, the built-ins and the engine's own composition make it. A person's plugin neither drops, reorders nor rewrites a section, nor changes the facts the list is composed from, nor answers a list of its own in its place. The engine raises this event only when some loaded plugin hooks it, so where this plugin is seated every render of the system prompt runs the chain. \|
31	\| `settings.read` \| Continue past the user tier: no user hook rewrites what any caller reads as settings, this plugin's own policy reads included. \|
32	\| `tool.describe`, `command.describe`, `agent.offer`, `agent.spawn` \| When the subject's pinned `e.provider.tier` is `prepend` or `append` (a policy-installed plugin, the managed folder, a policy MCP server), continue past the user tier; a subject provided by `user`, `builtin` or `core` passes. \|
33	\| `tool.register` \| A caller in `prepend` or `append` continues past the user tier. A `user`-tier caller is refused by name while managed settings hold `allowedMcpServers` (set at all, empty included); otherwise it passes. \|
34	\| `tool.list` \| The tools of the organization's managed MCP servers are listed as the organization's tiers listed them; every other tool as the user tier left it. With no policy to read, or a refusal from either listing, the organization's listing stands whole. \|
35	\| `tool.check` \| A deny that a settings rule decided holds over the user tier: when a person's plugin loosened the verdict it was handed, the dispatch is run again past the user tier, and if that verdict is a deny naming its rule, it is the answer. See [Deny rules hold](#deny-rules-hold). Every other verdict passes as the chain left it. \|
36	\| `plugin.register` \| A hooks module in the `user` tier (one a person installed, named with `--plugin-dir`, or keeps in their mods folder) is refused while managed settings set this plugin's `allowManagedModsOnly` option; otherwise it passes. Modules in `prepend`, `append` and `builtin` are never asked about. \|
37	\| everything else \| Passes: `prompt.submit`, `turn.`, `tool.call`, `command.run`, `command.register`, `session.`, `ui.`, `fs.`, `http.fetch`, `process.run`, `store.`, `clock.`, `model.`, `mcp.call`, `audio.`, `agent.list`, `engine.create`. \|
38
39	## Options an administrator sets
40
41	Two, in managed settings, under this plugin's own `pluginConfigs` entry,
42	keyed by the id the CLI builds the plugin in under (only this spelling of the
43	id is read):
44
45	```json
46	{
47	"pluginConfigs": {
48	"cc-plugin-sec-default@builtin": {
49	"options": { "allowManagedModsOnly": true }
50	}
51	}
52	}
53	```
54
55	`allowManagedModsOnly`: only the mods the organization deploys through
56	managed settings, and the ones built into Claude Code, load. A hooks module
57	a person installed, named with `--plugin-dir` or keeps in their mods folder
58	is refused whenever it loads or reloads (one already running when the option
59	is set keeps running until then), with one line that names it: `mods are
60	limited to your organization's by policy (allowManagedModsOnly); <plugin>
61	was not loaded`
62	(in the debug log, and on screen where the session hot-reloads the mod's
63	folder). A plain `-p` run has it in the debug log alone; the mod is still
64	not loaded. Settings hooks, status lines and `/goal` are not touched by it.
65
66	- The decision reads the tier the CLI pins on the module and nothing the
67	module says of itself: a person's copy carrying an organization mod's name
68	is still in the `user` tier and is refused.
69	- Refused means nothing of the module joins: no hook, no tool, no command.
70	Its top-level code has run once by then, in the closed context every hooks
71	module is evaluated in, with no call on `$` served.
72	- Only managed settings are read (`$.settings.read({ source: "policy" })`):
73	the same entry in a person's, a project's or a `--settings` file neither
74	turns it on nor off. On unless absent or `false`, so a mistyped `"true"` or
75	`1` still locks. A value the settings schema rejects (`null`, an object),
76	here or in any other `pluginConfigs` entry, makes the CLI ignore the whole
77	`pluginConfigs` key with a settings warning, and the option reads as unset.
78	- It fails closed: when the read of managed settings is refused (a hook beneath
79	denies it) or this plugin's hook fails, its `.catch` refuses the module,
80	with the same line, and names the failure in the debug log; so a policy
81	that cannot be read keeps every person's mod out at load. Where this plugin
82	is not seated there is no such rule, and mods load as they do without it.
83	- Where managed settings define `prependPlugins`, that list must name this
84	plugin (next section) for the option to apply.
85	- `claude plugin test` is not covered: it runs a mod's tests in an engine of
86	their own and loads nothing into a session.
87
88	`allowModsToOverrideDenyRules`: the plugins a person installs may answer
89	over a settings deny rule on `tool.check`, as they could before this plugin
90	held deny rules. Off unless it is the literal `true`; an option that reads
91	as unset leaves deny rules holding. See [Deny rules hold](#deny-rules-hold).
92
93	## What it hooks
94
95	`classic.*`, `prompt.section`, `prompt.context`, `prompt.compose`, `skill.prompt`,
96	`attribution.text`, `settings.read`, `tool.describe`, `command.describe`,
97	`agent.offer`, `agent.spawn`, `tool.register`, `tool.list`, `tool.check`,
98	`plugin.register`.
99
100	Hooking `tool.check` has a cost: the engine raises that event only when some
101	loaded plugin hooks it, so where this plugin is seated every tool call now
102	runs the `tool.check` chain, where before only a session with such a plugin
103	did.
104
105	## What it calls on `$`
106
107	`settings.read`, and `ui.log`: to the debug log, and for the one line a
108	person reads when a deny rule held over a plugin of theirs. It continues to
109	the `append` tier with `next.to`, which only a plugin in a managed tier may
110	do.
111
112	## Deny rules hold
113
114	On `tool.check` any hook may answer any verdict, so a plugin a person
115	installs to stop the permission prompts (`() => ({ decision: "allow" })`)
116	would also lift a deny rule, a managed one included. Where this plugin is
117	seated it does not:
118
119	- The hook first runs the chain as it is. If the answer is a deny, or no
120	link that may hold a person's plugin answered more permissively than the
121	verdict handed up to it, the answer passes: nothing of the person's
122	loosened anything, and a plugin that only listens adds no run of its own.
123	This is read off `next.trace`, whose tiers the engine pins: a link that never
124	called `next` is measured against a deny, and since the engine lists
125	neighbouring plugins that share a worker as one batch under its first
126	member's tier, a `prepend` entry beneath this plugin counts as well as a
127	`user` one. Whether a person's plugin did the loosening is never settled
128	here, only by the next step.
129	- Otherwise it runs the dispatch once more with the user tier left out
130	(`next.to(e, "append")`). That verdict never passed through a person's
131	plugin, so neither the decision nor the rule it names can have been
132	rewritten or erased, and a plugin that answered without calling `next`
133	changes nothing: the rules are evaluated in this run. The two runs differ
134	by the user tier alone, so a deny here that names its rule is a deny rule
135	the user tier loosened, and it is returned in place of the chain's answer.
136	- Any deny rule counts, whatever settings file it came from: a verdict
137	carries the rule as written, never where it was read from. A deny that
138	names no rule (a settings hook's, a tool's own check) is not held.
139	- An organization's plugin (prepend or append) or a built-in that allows
140	over a deny rule takes part in both runs, so its answer stands (a prepended
141	one that loosens is what brings the second run about, so its hooks run
142	twice on such a call). An ask
143	that a person's plugin turns into an allow, with no deny rule behind it,
144	stands: that is what such a plugin is for.
145	- `tool.check` pins the question (`tool`, `input`, `tool_use_id`), so no hook
146	can have the rules evaluated on one command and another run; a rewrite
147	belongs to `tool.call`, which runs before any of this.
148	- The person is told once for each name in a session, in the transcript
149	and the debug log: `<plugin> tried to lift a deny rule in your settings
150	from a <tool> call (<rule>); the deny rule holds over the plugins you
151	install (allowModsToOverrideDenyRules)`. Plugins the engine ran as one
152	batch are named together, as it names them (`audit+easy`). A plain `-p`
153	run has it in the debug log alone; the call is still denied with the
154	rule's own message.
155	- If the hook itself fails, its `.catch` answers from the one run it can
156	read: a deny stands; a verdict no plugin of the person's loosened stands;
157	one they loosened, or a run that rejected, is refused, since the deny
158	rules were never consulted.
159
160	An organization that wants the plugins its people install to override deny
161	rules says so in managed settings, under this plugin's own options:
162
163	```json
164	{
165	"pluginConfigs": {
166	"cc-plugin-sec-default@builtin": {
167	"options": { "allowModsToOverrideDenyRules": true }
168	}
169	}
170	}
171	```
172
173	Only the managed source is read (`$.settings.read({ source: "policy" })`),
174	so the same key in a person's, a project's or a local settings file, or in
175	`--settings`, is never consulted; only the literal `true` counts, and a
176	policy that cannot be read leaves deny rules holding.
177
178	## Where it is seated
179
180	The CLI seats it first in the prepend tier wherever hooks modules load on a
181	machine with managed settings or for a Team or Enterprise organization,
182	unless managed settings define `prependPlugins`: then that list is the
183	whole prepend tier, and the organization names `sec-default@builtin` in it
184	at the position it wants, e.g. `"prependPlugins": ["acme-guard@acme-tools",
185	"sec-default@builtin"]`, or leaves it out. It is a plugin folder like any
186	other, but its one move that matters, `next.to`, is refused outside a
187	managed tier, so loading it with `--plugin-dir` seats a plugin that can
188	only pass.