3tears-agent-skills

Procedural memory for 3tears-based agents. Stores per-agent / per-user labeled markdown procedures (body) and tool-surface modifications (tool_additions, tool_restrictions) plus a prompt_mode enum controlling how the body interacts with the consumer's base system prompt.

This package provides the schema + Collection layer:

agent_skills -- one row per skill (partition column agent_id).
agent_skill_invocations -- one row per skill load with synchronous outcome tracking populated by the consumer's post-LLM hook (partition column agent_id; composite FK CASCADE on parent skill).
Two BaseEntity subclasses (AgentSkillEntity, AgentSkillInvocationEntity).
Two BaseCollection subclasses exposing the public skill-registry API.
A trigger-maintained search_vector (FTS for skill_list query-filter ranking; NOT auto-load).

Agent tools and the per-turn composition renderer do not live here.

Migration registration

from threetears.agent.skills import register as register_skills
from threetears.core.data.migrations import MigrationRunner

runner = MigrationRunner()
register_skills(runner)

Migrations are agent-scoped and depend on conversations (the invocation rows carry conversation_id -- ordering on apply, not an FK constraint).

`SkillRegistryClient` Protocol -- why this package takes no ACL / tools deps

agent-memory (a sibling package) takes direct dependencies on 3tears-agent-acl and 3tears-agent-tools because its memory tools need first-class ACL evaluator + tool-registry types in their public surface. agent-skills deliberately diverges. The tool factories take a thin SkillRegistryClient Protocol (three async methods, acl_permits / list_skill_eligible_tools / get_tool_introspect, plus a ConversationIdResolver / ActiveSkillProbe / ActiveSkillSetter callable triple wired by the consumer) instead of importing the ACL evaluator or the tool registry types directly.

The trade-off: callers must implement a small adapter (~10 lines over their existing NamespaceCollection + ACL cache + in-process tool registry), but agent-skills ships with zero hard deps beyond core and langchain-core. The consumer wires this; tests mock it. The ACL-via-evaluator approach targets the composition renderer, not the tools surface.

Future sibling packages should follow whichever pattern their public surface demands -- direct deps when types are part of the contract, Protocol when the contract is method-shaped and the deps are incidental.

Release files for 3tears-agent-skills 0.48.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for 3tears-agent-skills 0.48.0
File	Size	Uploaded
3tears_agent_skills-0.48.0.tar.gz	84.4 kB	Sep 22, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for 3tears-agent-skills 0.48.0
File	Interpreter	ABI	Platform	Reset
3tears_agent_skills-0.48.0-py3-none-any.whl 50.0 kB Sep 22, 2026	Python 3	none	any	Details

Total release size: 134.4 kB

Release files / 3tears_agent_skills-0.48.0.tar.gz

Download URL	3tears_agent_skills-0.48.0.tar.gz
Size	84.4 kB
Tags	Source
SHA-256 checksum How to use checksums	`f78cec7a36523d41719771fc1ab0f167ac0042345d574db637e6147896b87aab`
BLAKE2b-256 checksum How to use checksums	`35c0445c54f286e3e151ff2b508b422dd6231852d531f873f193b32093c5a06e`
Upload date	Sep 22, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Identity

Publishing platform: GitHub Actions
Publishing repository: github.com/pacepace/3tears
Publishing commit: github.com/pacepace/3tears/tree/b4e88790e510b8c77f7d4981732ae878ffd98a25

Workflow

Publishing configuration: github.com/pacepace/3tears/blob/b4e88790e510b8c77f7d4981732ae878ffd98a25/.github/workflows/release.yml
Publishing logs: github.com/pacepace/3tears/actions/runs/35663908880/attempts/1

Artifact

Subject: 3tears_agent_skills-0.48.0.tar.gz
SHA-256 checksum: f78cec7a36523d41719771fc1ab0f167ac0042345d574db637e6147896b87aab Verifying attestations

Release files / 3tears_agent_skills-0.48.0-py3-none-any.whl

Download URL	3tears_agent_skills-0.48.0-py3-none-any.whl
Size	50.0 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`17610d8c76d71ab170737dba14e827646d17f473ae86aa81a310fdb9bf932747`
BLAKE2b-256 checksum How to use checksums	`ac88b7fd6525f83bf8e37f0321eb19f7899a27319c4da9673c2f201dd787ff8e`
Upload date	Sep 22, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`twine/7.0.0 CPython/3.13.14`

Provenance

Identity

Publishing platform: GitHub Actions
Publishing repository: github.com/pacepace/3tears
Publishing commit: github.com/pacepace/3tears/tree/b4e88790e510b8c77f7d4981732ae878ffd98a25

Workflow

Publishing configuration: github.com/pacepace/3tears/blob/b4e88790e510b8c77f7d4981732ae878ffd98a25/.github/workflows/release.yml
Publishing logs: github.com/pacepace/3tears/actions/runs/35663908880/attempts/1

Artifact

Subject: 3tears_agent_skills-0.48.0-py3-none-any.whl
SHA-256 checksum: 17610d8c76d71ab170737dba14e827646d17f473ae86aa81a310fdb9bf932747 Verifying attestations