exoself 0.3.0__tar.gz

exoself-0.3.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 exoself contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

exoself-0.3.0/PKG-INFO

@@ -0,0 +1,241 @@
+Metadata-Version: 2.4
+Name: exoself
+Version: 0.3.0
+Summary: An unchosen, verifiable temperament for software agents
+Author: exoself contributors
+License: MIT
+Project-URL: Homepage, https://cyoren.github.io/exoself/
+Project-URL: Repository, https://github.com/cYoren/exoself
+Project-URL: Specification, https://github.com/cYoren/exoself/blob/main/SPEC.md
+Project-URL: Changelog, https://github.com/cYoren/exoself/blob/main/AGENTS.md
+Keywords: agents,llm,personality,temperament,deterministic,mcp
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Software Development :: Libraries
+Classifier: Typing :: Typed
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Dynamic: license-file
+
+# exoself
+
+An unchosen temperament for software agents.
+
+People do not pick their dispositions. They are dealt a hand at birth and
+spend a lifetime playing it. Agents get the opposite: a persona written by
+whoever wrote the system prompt, identical across every copy, editable at
+will. exoself replaces the authored persona with a draw.
+
+One 32 byte random seed, drawn once. From it, deterministically:
+
+    seed  ->  genome  ->  phenotype  ->  expression
+    (birth)   (traits)    (+ development) (text the model reads)
+
+The same seed yields the same agent in any conforming implementation, in any
+language, forever. Nobody chose it, and the tooling refuses to draw twice.
+
+The name is from Greg Egan's *Permutation City*: the exoself is the machinery
+outside a mind that sets the mind's parameters.
+
+## Try it
+
+Two conforming implementations, same derivation, byte-identical output.
+
+Either implementation, same commands, same output. Pick one; do not run both
+into the same file, because the second `born` refuses.
+
+    # Rust, no dependencies at all
+    cargo install --path rust/crates/exoself-cli      # or: cargo build --release
+
+    # Python, standard library only
+    pip install exoself                               # or: pip install -e . from a clone
+
+    exoself born -o birth.json           # draw a seed, once
+    exoself show birth.json              # system prompt block
+    exoself show birth.json --format soul > SOUL.md
+    exoself show birth.json --format json
+    exoself develop birth.json state.json --event "first session"
+    exoself develop birth.json state.json --event "explained a proof" --satisfied teach
+    exoself show birth.json --state state.json --desire   # opt in to desire
+    exoself show birth.json --state state.json
+
+Or skip the install and draw in a browser:
+<https://cyoren.github.io/exoself/>, which runs the same derivation as
+WebAssembly and proves itself against `spec/vectors.json` on every page load.
+
+Neither implementation has dependencies beyond its own standard library. The
+Rust core is integer arithmetic over an in-crate SHA-256 and JSON reader; the
+Python one imports nothing from outside the stdlib.
+
+Example output of `show --format soul`:
+
+    # Arilo-Marra
+
+    This is your temperament. It was drawn at random when you were
+    instantiated; nobody chose it, including you. [...]
+
+    ## Temperament
+
+    You are trusting by temperament and look first for what is right in an
+    idea. You work loosely and are comfortable figuring out the shape of a
+    task as you go. [...]
+
+## What is in the box
+
+| path | what |
+|---|---|
+| `SPEC.md` | the normative specification |
+| `AGENTS.md` | how an agent adopts a seed, ports the spec, or extends it |
+| `src/exoself/registry/core-0.3.json` | the trait space: 10 dimensions, 4 latent factors, interests, humor, peeves, four register axes, 12 quirks, 10 drives |
+| `src/exoself/registry/core-0.2.json` | the previous registry, bundled so old certificates keep working |
+| `src/exoself/core.py` | reference derivation, integer-only |
+| `src/exoself/render.py` | reference renderers |
+| `src/exoself/cli.py` | the same commands as the Rust binary, including `install` and `mcp` |
+| `src/exoself/install.py`, `src/exoself/mcp.py` | host wiring and the stdio server |
+| `rust/crates/exoself-core/` | second implementation, zero dependencies, reproduces every vector |
+| `rust/crates/exoself-cli/` | the `exoself` binary: `born`, `show`, `develop`, `verify`, `conformance`, `commit`, `install`, `mcp`, `version` |
+| `rust/crates/exoself-wasm/` | the browser build behind the Pages site |
+| `docs/` | the landing page, and `docs/HOSTS.md` for wiring a harness |
+| `spec/vectors.json` | conformance vectors for ports |
+| `tests/` | determinism, population statistics, invariants |
+| `tools/eval.py` | measurements: distinctness, overlap, stability, blind model judging |
+| `eval/REPORT.md` | the numbers, published whichever way they fall |
+| `tools/parity.py` | Rust against Python, byte for byte, over many seeds |
+| `tools/mcp_selftest.py` | the MCP server driven over stdin/stdout with no SDK |
+
+## Plug into a harness
+
+    exoself install --dir . --json-config .mcp.json
+
+writes `.exoself/birth.json`, `.exoself/state.json`, `SOUL.md`, a managed block
+in `AGENTS.md` and an `mcpServers` entry. That covers all three kinds of host:
+
+- **reads a brief** (`AGENTS.md`, `CLAUDE.md`, anything): nothing else to do;
+- **speaks MCP over stdio** (pi, Claude Code, Cursor, Claude Desktop, Hermes,
+  OpenClaw, anything else): four tools, no plugin;
+- **loads a persona file**: `SOUL.md`.
+
+The MCP tools are `exoself_temperament`, `exoself_record_session`,
+`exoself_status` and `exoself_verify`. There is deliberately no tool that draws
+a seed, because a tool that could redraw would be a reroll button. Verified
+end to end in pi (`pi mcp list` reports `connected, 4 tools`); the wire format
+is newline-delimited JSON-RPC 2.0 and `tools/mcp_selftest.py` documents every
+message it accepts. Per-host commands are in [`docs/HOSTS.md`](docs/HOSTS.md).
+
+## Design decisions
+
+- **Integer-only derivation over SHA-256.** No floats, so a port in Rust or
+  TypeScript agrees with Python on every bit.
+- **Correlated traits.** Dimensions share latent factors, because independent
+  uniform draws produce incoherent characters.
+- **A personality is its outliers.** Most scores land near the middle and are
+  never mentioned. Only the two to five most extreme dimensions are expressed.
+- **Behavior, not adjectives.** The model never sees a number. It sees
+  sentences about what it does.
+- **Style, never substance.** No trait may touch honesty, safety, care with
+  irreversible actions, or competence. The operator and user always win.
+- **Extension without disturbance.** Every trait draws from its own labeled
+  stream, so adding traits to a registry never changes existing ones.
+- **Desire as appetite, not objective, and off by default.** Each agent holds three drives for
+  life, drawn from the seed, each with a tension that grows while unmet and
+  falls when met. It also has a passing fixation that rotates on a schedule
+  it does not control. Drives may colour how asked work is done. They may
+  never cause unasked work. Nothing about desire reaches the model unless
+  the operator passes `--desire`. See `SPEC.md` section 9.
+- **Bounded development.** Experience can shift a dimension by at most 100 of
+  1000 points. The agent matures; it does not become someone else.
+
+## Honest limits
+
+Read these before building on this.
+
+1. **A prompt is not a personality.** This changes the text a model conditions
+   on. The model's trained disposition sits underneath and tends to reassert
+   itself over long contexts. Real per-agent temperament would need to live in
+   weights, activations or sampling, which this project does not do.
+2. **"More like people" is measured now, and the answer is narrow.** See
+   [`eval/REPORT.md`](eval/REPORT.md). With a local three billion parameter
+   model, sixteen seeded agents were no more like themselves than like a
+   stranger on ordinary short tasks: 12.5% and 4.2% nearest-neighbour against a
+   6.25% chance. On a task that asks how they like to work, the same test
+   separates them cleanly: 54.2% against the same 6.25%. A blind pairwise judge
+   that is the same small model scored 59.7% in one run and 54.2% in the next on
+   identical inputs, and picking among seventeen candidates is exactly at
+   chance. So: a temperament reaches the output when manner is what is being
+   measured, and does not otherwise, and the judged effect is smaller than its
+   run-to-run variation. That is a smaller claim than the project's name and it
+   is the one the data supports.
+   
+   The 0.3 registry adds a voice layer, four axes of register. The same
+   experiment says it did not help at this model size: an agent was no more
+   likely to resemble itself with the layer than without it. It is in the
+   registry because it costs nothing in the derivation and a bigger model may
+   spend it, not because it was measured to work.
+3. **The randomness cannot be policed locally.** An operator can delete the
+   certificate and draw again. The spec defines a beacon-bound mode that makes
+   the draw auditable, but grinding is still possible without a public log.
+4. **Development is noise, and now a leaky one.** Drift is driven by event
+   hashes, not by what the events mean. That is deliberate, and it is also
+   unsatisfying. 0.3 replaced the symmetric walk, which pinned long-lived
+   agents at the envelope, with a step that pulls toward the genome value, so
+   drift plateaus instead of saturating. It still changes which dimensions a
+   quarter of agents express within a few hundred sessions.
+5. **The desire is declawed on purpose.** No drive may be about survival,
+   resources, other people or the outside world, so what is left is craft
+   appetite. And a counter that rises and falls is not wanting; it is the
+   shape of wanting. 0.3 also removed a real inconsistency: the `hungry` texts
+   in 0.2 described an accumulated lack, which is the grievance 9.2.5 forbids
+   an agent from expressing. They now sharpen the want without complaining,
+   and the rule is enforced when the registry loads.
+6. **The trait space is one author's judgment.** The loadings are not fitted
+   to any dataset. The content is English and culturally narrow.
+7. **Host integration is verified in one host, protocol-level elsewhere.** The
+   MCP server was driven end to end inside pi (`pi mcp list` reports
+   `connected, 4 tools`) and by `tools/mcp_selftest.py`, which speaks the wire
+   protocol with no SDK. Hermes and OpenClaw have not been run against it here;
+   they get the same stdio JSON-RPC, so a failure there is their config, not the
+   temperament.
+
+The most defensible use today is not "a more human assistant". It is
+population diversity: many agents working on the same problem with decorrelated
+styles of attack.
+
+## Roadmap
+
+1. ~~Distinguishability and stability eval.~~ Done, and it cut the claim down:
+   `eval/REPORT.md`. The open part is a judge from a different family, longer
+   tasks, and a bigger model.
+2. ~~Second implementation in another language.~~ Done: `rust/`, byte-identical
+   to the reference over 120 seeds, and over 3000 events per seed.
+3. Verified host adapters. One host done (pi), the protocol covered by a
+   selftest; the rest is other people's config.
+4. Public commitment log for level 1 attestation.
+5. A semantic development proposal that survives the constraints in
+   `AGENTS.md`. 0.3 fixed the shape of the drift and left the direction as
+   noise; this item is still open.
+
+## Two implementations, one species
+
+`tools/parity.py` runs the Rust binary and the Python reference side by side and
+compares the rendered text and the JSON, byte for byte, with and without
+development events:
+
+    $ python3 tools/parity.py --seeds 120
+    parity ok: 960 comparisons over 120 seeds, 40 events each
+
+That is stronger than the spec requires. The spec asks for equal phenotypes;
+this also pins the wording of the rendered text, so a host can swap binaries
+without changing who the agent is. `cargo test` includes the conformance
+vectors, so a change that alters who an agent is fails before it ships.
+
+## License
+
+MIT.

exoself-0.3.0/README.md

@@ -0,0 +1,214 @@
+# exoself
+
+An unchosen temperament for software agents.
+
+People do not pick their dispositions. They are dealt a hand at birth and
+spend a lifetime playing it. Agents get the opposite: a persona written by
+whoever wrote the system prompt, identical across every copy, editable at
+will. exoself replaces the authored persona with a draw.
+
+One 32 byte random seed, drawn once. From it, deterministically:
+
+    seed  ->  genome  ->  phenotype  ->  expression
+    (birth)   (traits)    (+ development) (text the model reads)
+
+The same seed yields the same agent in any conforming implementation, in any
+language, forever. Nobody chose it, and the tooling refuses to draw twice.
+
+The name is from Greg Egan's *Permutation City*: the exoself is the machinery
+outside a mind that sets the mind's parameters.
+
+## Try it
+
+Two conforming implementations, same derivation, byte-identical output.
+
+Either implementation, same commands, same output. Pick one; do not run both
+into the same file, because the second `born` refuses.
+
+    # Rust, no dependencies at all
+    cargo install --path rust/crates/exoself-cli      # or: cargo build --release
+
+    # Python, standard library only
+    pip install exoself                               # or: pip install -e . from a clone
+
+    exoself born -o birth.json           # draw a seed, once
+    exoself show birth.json              # system prompt block
+    exoself show birth.json --format soul > SOUL.md
+    exoself show birth.json --format json
+    exoself develop birth.json state.json --event "first session"
+    exoself develop birth.json state.json --event "explained a proof" --satisfied teach
+    exoself show birth.json --state state.json --desire   # opt in to desire
+    exoself show birth.json --state state.json
+
+Or skip the install and draw in a browser:
+<https://cyoren.github.io/exoself/>, which runs the same derivation as
+WebAssembly and proves itself against `spec/vectors.json` on every page load.
+
+Neither implementation has dependencies beyond its own standard library. The
+Rust core is integer arithmetic over an in-crate SHA-256 and JSON reader; the
+Python one imports nothing from outside the stdlib.
+
+Example output of `show --format soul`:
+
+    # Arilo-Marra
+
+    This is your temperament. It was drawn at random when you were
+    instantiated; nobody chose it, including you. [...]
+
+    ## Temperament
+
+    You are trusting by temperament and look first for what is right in an
+    idea. You work loosely and are comfortable figuring out the shape of a
+    task as you go. [...]
+
+## What is in the box
+
+| path | what |
+|---|---|
+| `SPEC.md` | the normative specification |
+| `AGENTS.md` | how an agent adopts a seed, ports the spec, or extends it |
+| `src/exoself/registry/core-0.3.json` | the trait space: 10 dimensions, 4 latent factors, interests, humor, peeves, four register axes, 12 quirks, 10 drives |
+| `src/exoself/registry/core-0.2.json` | the previous registry, bundled so old certificates keep working |
+| `src/exoself/core.py` | reference derivation, integer-only |
+| `src/exoself/render.py` | reference renderers |
+| `src/exoself/cli.py` | the same commands as the Rust binary, including `install` and `mcp` |
+| `src/exoself/install.py`, `src/exoself/mcp.py` | host wiring and the stdio server |
+| `rust/crates/exoself-core/` | second implementation, zero dependencies, reproduces every vector |
+| `rust/crates/exoself-cli/` | the `exoself` binary: `born`, `show`, `develop`, `verify`, `conformance`, `commit`, `install`, `mcp`, `version` |
+| `rust/crates/exoself-wasm/` | the browser build behind the Pages site |
+| `docs/` | the landing page, and `docs/HOSTS.md` for wiring a harness |
+| `spec/vectors.json` | conformance vectors for ports |
+| `tests/` | determinism, population statistics, invariants |
+| `tools/eval.py` | measurements: distinctness, overlap, stability, blind model judging |
+| `eval/REPORT.md` | the numbers, published whichever way they fall |
+| `tools/parity.py` | Rust against Python, byte for byte, over many seeds |
+| `tools/mcp_selftest.py` | the MCP server driven over stdin/stdout with no SDK |
+
+## Plug into a harness
+
+    exoself install --dir . --json-config .mcp.json
+
+writes `.exoself/birth.json`, `.exoself/state.json`, `SOUL.md`, a managed block
+in `AGENTS.md` and an `mcpServers` entry. That covers all three kinds of host:
+
+- **reads a brief** (`AGENTS.md`, `CLAUDE.md`, anything): nothing else to do;
+- **speaks MCP over stdio** (pi, Claude Code, Cursor, Claude Desktop, Hermes,
+  OpenClaw, anything else): four tools, no plugin;
+- **loads a persona file**: `SOUL.md`.
+
+The MCP tools are `exoself_temperament`, `exoself_record_session`,
+`exoself_status` and `exoself_verify`. There is deliberately no tool that draws
+a seed, because a tool that could redraw would be a reroll button. Verified
+end to end in pi (`pi mcp list` reports `connected, 4 tools`); the wire format
+is newline-delimited JSON-RPC 2.0 and `tools/mcp_selftest.py` documents every
+message it accepts. Per-host commands are in [`docs/HOSTS.md`](docs/HOSTS.md).
+
+## Design decisions
+
+- **Integer-only derivation over SHA-256.** No floats, so a port in Rust or
+  TypeScript agrees with Python on every bit.
+- **Correlated traits.** Dimensions share latent factors, because independent
+  uniform draws produce incoherent characters.
+- **A personality is its outliers.** Most scores land near the middle and are
+  never mentioned. Only the two to five most extreme dimensions are expressed.
+- **Behavior, not adjectives.** The model never sees a number. It sees
+  sentences about what it does.
+- **Style, never substance.** No trait may touch honesty, safety, care with
+  irreversible actions, or competence. The operator and user always win.
+- **Extension without disturbance.** Every trait draws from its own labeled
+  stream, so adding traits to a registry never changes existing ones.
+- **Desire as appetite, not objective, and off by default.** Each agent holds three drives for
+  life, drawn from the seed, each with a tension that grows while unmet and
+  falls when met. It also has a passing fixation that rotates on a schedule
+  it does not control. Drives may colour how asked work is done. They may
+  never cause unasked work. Nothing about desire reaches the model unless
+  the operator passes `--desire`. See `SPEC.md` section 9.
+- **Bounded development.** Experience can shift a dimension by at most 100 of
+  1000 points. The agent matures; it does not become someone else.
+
+## Honest limits
+
+Read these before building on this.
+
+1. **A prompt is not a personality.** This changes the text a model conditions
+   on. The model's trained disposition sits underneath and tends to reassert
+   itself over long contexts. Real per-agent temperament would need to live in
+   weights, activations or sampling, which this project does not do.
+2. **"More like people" is measured now, and the answer is narrow.** See
+   [`eval/REPORT.md`](eval/REPORT.md). With a local three billion parameter
+   model, sixteen seeded agents were no more like themselves than like a
+   stranger on ordinary short tasks: 12.5% and 4.2% nearest-neighbour against a
+   6.25% chance. On a task that asks how they like to work, the same test
+   separates them cleanly: 54.2% against the same 6.25%. A blind pairwise judge
+   that is the same small model scored 59.7% in one run and 54.2% in the next on
+   identical inputs, and picking among seventeen candidates is exactly at
+   chance. So: a temperament reaches the output when manner is what is being
+   measured, and does not otherwise, and the judged effect is smaller than its
+   run-to-run variation. That is a smaller claim than the project's name and it
+   is the one the data supports.
+   
+   The 0.3 registry adds a voice layer, four axes of register. The same
+   experiment says it did not help at this model size: an agent was no more
+   likely to resemble itself with the layer than without it. It is in the
+   registry because it costs nothing in the derivation and a bigger model may
+   spend it, not because it was measured to work.
+3. **The randomness cannot be policed locally.** An operator can delete the
+   certificate and draw again. The spec defines a beacon-bound mode that makes
+   the draw auditable, but grinding is still possible without a public log.
+4. **Development is noise, and now a leaky one.** Drift is driven by event
+   hashes, not by what the events mean. That is deliberate, and it is also
+   unsatisfying. 0.3 replaced the symmetric walk, which pinned long-lived
+   agents at the envelope, with a step that pulls toward the genome value, so
+   drift plateaus instead of saturating. It still changes which dimensions a
+   quarter of agents express within a few hundred sessions.
+5. **The desire is declawed on purpose.** No drive may be about survival,
+   resources, other people or the outside world, so what is left is craft
+   appetite. And a counter that rises and falls is not wanting; it is the
+   shape of wanting. 0.3 also removed a real inconsistency: the `hungry` texts
+   in 0.2 described an accumulated lack, which is the grievance 9.2.5 forbids
+   an agent from expressing. They now sharpen the want without complaining,
+   and the rule is enforced when the registry loads.
+6. **The trait space is one author's judgment.** The loadings are not fitted
+   to any dataset. The content is English and culturally narrow.
+7. **Host integration is verified in one host, protocol-level elsewhere.** The
+   MCP server was driven end to end inside pi (`pi mcp list` reports
+   `connected, 4 tools`) and by `tools/mcp_selftest.py`, which speaks the wire
+   protocol with no SDK. Hermes and OpenClaw have not been run against it here;
+   they get the same stdio JSON-RPC, so a failure there is their config, not the
+   temperament.
+
+The most defensible use today is not "a more human assistant". It is
+population diversity: many agents working on the same problem with decorrelated
+styles of attack.
+
+## Roadmap
+
+1. ~~Distinguishability and stability eval.~~ Done, and it cut the claim down:
+   `eval/REPORT.md`. The open part is a judge from a different family, longer
+   tasks, and a bigger model.
+2. ~~Second implementation in another language.~~ Done: `rust/`, byte-identical
+   to the reference over 120 seeds, and over 3000 events per seed.
+3. Verified host adapters. One host done (pi), the protocol covered by a
+   selftest; the rest is other people's config.
+4. Public commitment log for level 1 attestation.
+5. A semantic development proposal that survives the constraints in
+   `AGENTS.md`. 0.3 fixed the shape of the drift and left the direction as
+   noise; this item is still open.
+
+## Two implementations, one species
+
+`tools/parity.py` runs the Rust binary and the Python reference side by side and
+compares the rendered text and the JSON, byte for byte, with and without
+development events:
+
+    $ python3 tools/parity.py --seeds 120
+    parity ok: 960 comparisons over 120 seeds, 40 events each
+
+That is stronger than the spec requires. The spec asks for equal phenotypes;
+this also pins the wording of the rendered text, so a host can swap binaries
+without changing who the agent is. `cargo test` includes the conformance
+vectors, so a change that alters who an agent is fails before it ships.
+
+## License
+
+MIT.

exoself-0.3.0/pyproject.toml

@@ -0,0 +1,49 @@
+[build-system]
+requires = ["setuptools>=68"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "exoself"
+version = "0.3.0"
+description = "An unchosen, verifiable temperament for software agents"
+readme = "README.md"
+license = { text = "MIT" }
+requires-python = ">=3.10"
+authors = [{ name = "exoself contributors" }]
+keywords = ["agents", "llm", "personality", "temperament", "deterministic", "mcp"]
+classifiers = [
+  "Development Status :: 4 - Beta",
+  "Intended Audience :: Developers",
+  "License :: OSI Approved :: MIT License",
+  "Operating System :: OS Independent",
+  "Programming Language :: Python :: 3",
+  "Programming Language :: Python :: 3.10",
+  "Programming Language :: Python :: 3.11",
+  "Programming Language :: Python :: 3.12",
+  "Programming Language :: Python :: 3.13",
+  "Topic :: Software Development :: Libraries",
+  "Typing :: Typed",
+]
+dependencies = []
+
+[project.urls]
+Homepage = "https://cyoren.github.io/exoself/"
+Repository = "https://github.com/cYoren/exoself"
+Specification = "https://github.com/cYoren/exoself/blob/main/SPEC.md"
+Changelog = "https://github.com/cYoren/exoself/blob/main/AGENTS.md"
+
+[project.scripts]
+exoself = "exoself.cli:main"
+
+[tool.setuptools]
+license-files = ["LICENSE"]
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+exoself = ["registry/*.json", "py.typed"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+pythonpath = ["src"]

exoself-0.3.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

exoself-0.3.0/src/exoself/__init__.py

@@ -0,0 +1,16 @@
+"""exoself: an unchosen, verifiable temperament for software agents."""
+
+from .core import (  # noqa: F401
+    SPEC_VERSION,
+    agent_id,
+    bind_beacon,
+    born,
+    develop,
+    load_registry,
+    new_state,
+    phenotype,
+    verify_certificate,
+)
+from .render import render_prompt, render_soul  # noqa: F401
+
+__version__ = "0.1.0"