security-boundary
Skill1 starsChanged 24 days agoMITView on GitHub
How every msmcp filesystem access must pass through the security boundary, and the error hierarchy that reports boundary violations.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- add-mcp-toolSkill
- data-referencesSkill
- instrument-class-defaultsSkill
- ms-data-ingestionSkill
- quality-gateSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
Use this for any code path that resolves a user-supplied path or a resource limit.
- Distinctive: line 56
Do not return a generic "analysis failed" string.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Security boundary
Use this for any code path that resolves a user-supplied path or a resource
limit. Every tool that touches the local filesystem must pass its paths
through msmcp.security. This includes acquisition files and spectral library
files alike; a library is not exempt.
The three hard limits
Enforced by SecurityPolicy (src/msmcp/security.py):
| Limit | Default | Override |
| Allowed root | process working directory | MSMCP_ALLOWED_ROOT |
| Max file size | 500 MiB | MSMCP_MAX_FILE_SIZE_BYTES |
| Max spectra per request | 1,000 | MSMCP_MAX_SPECTRA |
DEFAULT_POLICY reads the environment once at import time. Build a custom
policy with SecurityPolicy.from_env(**overrides) when a test or embedder needs
different limits; break-glass precedence is overrides, then environment, then
defaults.
Resolution rule
resolve_path makes the path absolute, expands ~, and resolves symlinks
before the containment check. A symlink inside the root that points outside
it is rejected, not followed. Never hand a raw user path to open().
Then call validate_file_size and validate_spectrum_count as appropriate.
Error hierarchy
Tools raise typed errors so the host can route the next step. Put a new error in the right place:
MsmcpError (errors.py)
├── InaccessiblePathError # missing, not a file, no permission
├── MalformedFileError # opened but does not parse
├── MissingDependencyError # optional package absent
├── UnsupportedFormatError # recognised but not supported
├── SpectrumIndexError # no spectrum at the requested index
└── SecurityError (security.py)
├── PathEscapeError # resolved outside the allowed root
├── FileSizeExceededError # exceeds the byte limit
└── SpectrumCountExceededErrorDo not return a generic "analysis failed" string. The point of the hierarchy is that the why survives to the caller.
Resource limits beyond files
The server-side store (state/store.py, state/pointers.py) has its own
bounds, enforced by refusing new registrations rather than evicting live
data — a resource error must never become a wrong answer:
MSMCP_MAX_REFERENCESMSMCP_MAX_TOTAL_BYTESMSMCP_REFERENCE_TTL_SECONDS(default one hour)
A StoreLimitError is the expected outcome when a limit is hit.
Posture reminders
- The server is offline and self-contained: no network, no credentials, no
hosted service.
open_world_hint=Falseon every tool. - There is no direct instrument control and no safety interlock layer in this repository, by design.
Tests
tests/test_security.py covers the boundary directly; tests/test_pointers.py
covers the store limits. When adding a limit or an error, add a case that
asserts the specific typed error, not just that something raised.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

