Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    tessera

    1

    Ask your agent to fix a bug, add a feature, write a test or explain a flow.

    Free

    Security scanned
    tessera

    tessera

    Example session with this skill installed

    Use the Tessera skill. In this Jinja checkout, where is UndefinedError raised for an undefined value and how is its message built? Show me the relevant code.

    • Read your context and instructions
    • Compiled the tessera

    Summary

    One call (uvx tessera-skill@1.0.0 pack "Undefined StrictUndefined _fail_with_undefined_error _undefined_message" --task lookup) on Jinja returned the relevant definitions as exact code in about 992 estimated tokens (budget 1500). The whole files that contain them are about 22,775 estimated tokens, so this example is about 96% smaller. The footer shows that the base class Undefined did not fit the budget.

    What the agent delivered (real, unedited tool output)

    PACK "Undefined StrictUndefined _fail_with_undefined_error _und..." confidence=HIGH budget=1500 (task lookup) seeds: Undefined._fail_with_undefined_error, StrictUndefined, TemplateError.message
    === FULL src/jinja2/exceptions.py::TemplateError.message L13-15
    
    ```python
        @property
        def message(self) -> str | None:
    ```
    
            return self.args[0] if self.args else None
    === FULL src/jinja2/parser.py::Parser.fail_unknown_tag L130-135
        def fail_unknown_tag(self, name: str, lineno: int | None = None) -> "te.NoReturn":
            """Called if the parser encounters an unknown tag.  Tries to fail
            with a human readable error message that could help to identify
            the problem.
            """
    
    ```python
            self._fail_ut_eof(name, self._end_token_stack, lineno)
    === FULL src/jinja2/runtime.py::Undefined._undefined_message L830-850
        @property
        def _undefined_message(self) -> str:
    ```
    
            """Build a message about the undefined value based on how it was
            accessed.
            """
    
    ```python
            if self._undefined_hint:
                return self._undefined_hint
    
            if self._undefined_obj is missing:
                return f"{self._undefined_name!r} is undefined"
    
            if not isinstance(self._undefined_name, str):
                return (
                    f"{object_type_repr(self._undefined_obj)} has no"
                    f" element {self._undefined_name!r}"
                )
    
            return (
                f"{object_type_repr(self._undefined_obj)!r} has no"
                f" attribute {self._undefined_name!r}"
            )
    === FULL src/jinja2/runtime.py::Undefined._fail_with_undefined_error L852-859
        @internalcode
        def _fail_with_undefined_error(
    ```
    
            self, *args: t.Any, **kwargs: t.Any
        ) -> "te.NoReturn":
            """Raise an :exc:`UndefinedError` when operations are performed
            on the undefined value.
            """
            raise self._undefined_exception(self._undefined_message)
    === FULL src/jinja2/runtime.py::StrictUndefined L1039-1062
    class StrictUndefined(Undefined):
        """An undefined that barks on print and iteration as well as boolean
        tests and all kinds of comparisons.  In other words: you can do nothing
        with it except checking if it's defined using the `defined` test.
    
        >>> foo = StrictUndefined(name='foo')
        >>> str(foo)
        Traceback (most recent call last):
          ...
        jinja2.exceptions.UndefinedError: 'foo' is undefined
        >>> not foo
        Traceback (most recent call last):
          ...
        jinja2.exceptions.UndefinedError: 'foo' is undefined
        >>> foo + 42
        Traceback (most recent call last):
          ...
        jinja2.exceptions.UndefinedError: 'foo' is undefined
        """
    
        __slots__ = ()
        __iter__ = __str__ = __len__ = Undefined._fail_with_undefined_error
        __eq__ = __ne__ = __bool__ = __hash__ = Undefined._fail_with_undefined_error
        __contains__ = Undefined._fail_with_undefined_error
    === FULL src/jinja2/utils.py::internalcode L98-101
    def internalcode(f: F) -> F:
        """Marks the function as internally used"""
    
    ```
        internal_code.add(f.__code__)
        return f
    ```
    
    OMITTED: src/jinja2/runtime.py::Undefined; src/jinja2/async_utils.py (outline); src/jinja2/exceptions.py (outline); src/jinja2/nodes.py (outline); src/jinja2/utils.py (outline); src/jinja2/environment.py (outline); src/jinja2/runtime.py (outline); src/jinja2/compiler.py (outline); src/jinja2/lexer.py (outline); src/jinja2/constants.py (outline); src/jinja2/defaults.py (outline); src/jinja2/parser.py (outline); src/jinja2/bccache.py (outline); src/jinja2/ext.py (outline); src/jinja2/loaders.py (outline); +1 more
    **WARNINGS:** skipped dir: .git/ (default ignore)
    **BUDGET-LIMITED:** 1 relevant items omitted (src/jinja2/runtime.py::Undefined). Rerun with --task test or --budget 2000, or use slice.py.
    **TOKENS:** ~992 / 1500 (estimate). Whole contributing files ~22775, saved ~96% (estimate)
    

    Answer the agent can now give. Undefined._fail_with_undefined_error raises UndefinedError with the text built by Undefined._undefined_message. StrictUndefined routes __str__, __iter__, __len__, __eq__, __bool__ and __hash__ to the same failing method.

    Next steps

    • Read the OMITTED line: Undefined did not fit, so run uvx tessera-skill@1.0.0 slice src/jinja2/runtime.py:Undefined --root <repo> if you need it.
    • Use the --task profile that matches the job (lookup, test, bug, explore, feature) to set the budget.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Give your AI agent only the code it needs.

    Tessera is a skill for Python repositories. Ask your agent to fix a bug, add a feature, write a test or explain a flow. Instead of opening whole files, it makes one call and gets a token-budgeted bundle: the best-matching functions and classes as exact code, outlines of related files, and a list of the items that did not fit.

    What your agent can do with it

    • pack: one call, a bundle of exact code plus outlines within a token budget (hard cap 8000, estimated)
    • find: keyword search for functions and classes
    • map: ranked overview of a repo
    • slice: outline of a file, an exact function, or a line range
    • deps: what a file imports and what imports it

    Measured (15 tasks on four Python repos: click, jinja, sqlparse, tinydb; full tables in BENCHMARK.md)

    agent-style keywordsplain keywords
    Token reduction vs reading whole files (median)69.4%70.1%
    Needed functions delivered as full code79.4%61.8%

    Requirements

    • uv installed once (https://docs.astral.sh/uv/), or pipx
    • Python 3.10 or newer
    • Internet access for the first call, which downloads the tool from PyPI; after that it works offline

    Standard library only. Read-only: the tool never writes files, reads environment variables or uses subprocesses. Python repositories only.

    Updates: subscribers receive improvements automatically; your agent picks up the newest version on its next call.

    Install: add the skill from Agensi, install uv once, then ask your agent to use the Tessera skill.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Code Quality & Review.

    • Python only; other languages are not supported.
    • The first call downloads the tool from PyPI, so it needs internet access once and uv (or pipx) installed.
    • About 8 of 10 needed functions arrive as full code with identifier-style keywords, about 6 of 10 with plain words. A missing function can be fetched with a follow-up call by its exact name.
    • A large function can be omitted or shown cut short when it does not fit the token budget; the OMITTED and TRUNCATED lines say so.
    • The confidence label (HIGH, MEDIUM, LOW) measures how well the keywords matched, not whether the bundle is complete.
    • Dynamic imports and runtime-generated code are not traced; callers are reported per file, not per function.
    • Token counts are estimates; each model counts differently.
    • Every call builds the index from scratch: on Django (2,937 files) one laptop took about 5 to 7 s per call when idle and 20 s or more for the first cold call; on a 91-file repo about 0.5 s.
    • Agents may not pick the skill by themselves: say "use the Tessera skill".

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 9 days ago

    • Free with an Agensi account

    Software
    uv (or pipx)Python 3.10 or newertessera-skill

    Permissions

    Read Files
    Network Access
    Terminal / Shell

    Runs a local command line tool (downloaded once from PyPI with uvx) that only reads .py files under the folder you pass as --root. Network is used only for that first download. The tool never writes files, reads environment variables or starts subprocesses. Secret-looking files (.env, *.pem, *.key) are never read.

    Listed9 days ago

    What's inside

    Frequently Asked Questions

    No. It runs locally with the Python standard library only. The only network use is the first download of the tool from PyPI.

    No. The tool only reads files.

    Python repositories for now. Future modifications would be coming.

    On the benchmark the bundle was a median of 69.4% to 70.1% smaller than reading the whole files that contain the target code. Check the OMITTED line at the end of each bundle to see what did not fit.

    Ask for it by its exact name, for example: uvx tessera-skill@1.0.0 pack "Class.method" --root <repo>. In the benchmark this found all 42 functions that a first call had missed.

    Subscribers get each new version of the skill automatically, with release notes by email. Your agent picks up the newest version on its next call.

    New in Code Quality & Review

    Free