Workspace Memory Skill

Workspace Memory Skill

Overview

This skill manages project memory inside the current workspace as Markdown files. It is for small-team workflows where agents and developers need shared, durable context without turning the repo into an unstructured dump of transcripts.

Prefer script-generated metadata. The agent should focus on semantic inputs and body content, not on hand-writing frontmatter.

Read [references/layout.md](references/layout.md) for the memory tree. Read [references/templates.md](references/templates.md) for schema and body-template navigation. Read [references/config.md](references/config.md) for memory-language configuration.

Routing

Pick the narrowest operation that matches the request:

• Initialize workspace memory:

Read [references/operations/initialize-memory.md](references/operations/initialize-memory.md)

• Record a new session:

Read [references/operations/session-capture.md](references/operations/session-capture.md)

• Compress recent memory into archive:

Read [references/operations/refine-recent.md](references/operations/refine-recent.md)

• Query existing memory:

Read [references/operations/query-memory.md](references/operations/query-memory.md)

• Maintain or deduplicate memory:

Read [references/maintenance.md](references/maintenance.md)

• Extract durable knowledge:

Read [references/operations/crystal-maintenance.md](references/operations/crystal-maintenance.md)

• Maintain topic summaries:

Read [references/operations/topic-summary-maintenance.md](references/operations/topic-summary-maintenance.md)

Core Rules

• Keep memory repo-local and Markdown-first.
• Treat structured metadata as system-owned fields. Generate or update them through scripts whenever possible.
• Treat body sections as content-owned fields. The agent should focus on the actual summary, rationale, decisions, and follow-ups.
• Prefer updating existing files over creating near-duplicates.
• Do not invent unsupported scripts. If an operation card says the flow is currently manual, follow the manual flow.
• Use script --help output as the final parameter contract when there is any ambiguity.
• Read docs/memory/config.toml before authoring body content when the memory tree exists.
• Keep filenames ASCII and keep memory headings in their scripted English form.
• Write body prose in the configured content_language. If the config is missing, default to English.

Current Automation Boundary

Script-supported today:

• initialization via scripts/init_memory.py
• session capture via scripts/record_session.py
• recent-to-archive compression via scripts/refine_memory.py
• query retrieval via scripts/query_memory.py
• crystal creation via scripts/create_crystal.py
• crystal updates via scripts/update_crystal.py
• topic-summary creation via scripts/create_topic_summary.py
• topic-summary updates via scripts/update_topic_summary.py
• language-policy validation via scripts/check_memory_language.py

Manual or partial today:

• broader maintenance and deduplication flows are still reference-driven
• broader derived-file freshness beyond recent/archive is still partial

Expected Outputs

Common outputs for this skill are:

• a new or updated memory directory under the current workspace
• one or more session files with script-generated metadata
• updated recent or archived summaries
• updated crystal or topic-summary files when the task requires durable knowledge extraction

Use file paths in your response when you create or update memory files.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: fightZy
• Source: fightZy/simple-skills
• License: MIT

Install and usage instructions live in the source repository linked above.