@mhrj/team-project-memory-skill
Advanced tools
@mhrj/team-project-memory-skill - npm Package Compare versions
+21
| MIT License | ||
| Copyright (c) 2026 MeharajM | ||
| 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. |
| interface: | ||
| display_name: "Team Memory Capture" | ||
| short_description: "Save a verified reusable learning" | ||
| default_prompt: "Use $team-memory-capture to draft a compact reusable project learning." | ||
| policy: | ||
| allow_implicit_invocation: false |
| --- | ||
| name: team-memory-capture | ||
| description: Capture a short reusable project learning after a non-obvious fix, failed approach, setup requirement, decision, or debugging path is verified and likely to save another agent time or tokens. | ||
| license: MIT | ||
| metadata: | ||
| author: meharajM | ||
| version: "0.2.0" | ||
| repository: meharajM/context-machine | ||
| --- | ||
| # Team Memory Capture | ||
| Capture only knowledge likely to prevent meaningful repeated work. | ||
| ## Before Writing | ||
| 1. Search with `$team-memory-search` for the same signature. | ||
| 2. Skip generic knowledge, obvious docs, raw transcripts/logs, temporary state, and duplicates. | ||
| 3. Redact secrets, tokens, personal paths, customer/personal data, private URLs, and exploit details. | ||
| 4. Keep the entry under 250 words; keep `Reuse` under 80. | ||
| ## Status | ||
| - `candidate`: plausible but not reproduced. | ||
| - `validated`: confirmed by test, CI, reproducible command, or user confirmation. | ||
| - `stale`: no longer preferred. | ||
| - `contradicted`: false for its stated scope. | ||
| Agent observation alone remains `candidate`. | ||
| ## Format | ||
| ```md | ||
| --- | ||
| id: <YYYYMMDD-HHMM-short-slug> | ||
| status: candidate | validated | stale | contradicted | ||
| confidence: low | medium | high | ||
| scope: <repo/path/component/tool/runtime> | ||
| signature: <normalized searchable signature> | ||
| environment: <important versions or none> | ||
| last_validated: <YYYY-MM-DD or unknown> | ||
| supersedes: [] | ||
| contradicts: [] | ||
| --- | ||
| # <Short title> | ||
| ## Reuse | ||
| <First action, prerequisites, and limits.> | ||
| ## Resolution | ||
| <Smallest useful cause and solution.> | ||
| ## Verify | ||
| `<command/check>` | ||
| ## Avoid | ||
| <Repeatable dead ends or none.> | ||
| ## Evidence | ||
| <Brief test, CI, PR, commit, or confirmation.> | ||
| ``` | ||
| ## Publish | ||
| Default to a draft. Do not commit, push, or open a PR without explicit permission or a repo policy that allows it. Prefer one file per learning. |
| interface: | ||
| display_name: "Team Memory Maintenance" | ||
| short_description: "Set up and clean shared memory" | ||
| default_prompt: "Use $team-memory-maintenance to configure or maintain the shared memory store." | ||
| policy: | ||
| allow_implicit_invocation: false |
| --- | ||
| name: team-memory-maintenance | ||
| description: Initialize or maintain a git-backed shared project-memory store, including policy, layout, deduplication, contradictions, staleness, retention, and conflict-safe publishing. | ||
| license: MIT | ||
| metadata: | ||
| author: meharajM | ||
| version: "0.2.0" | ||
| repository: meharajM/context-machine | ||
| --- | ||
| # Team Memory Maintenance | ||
| Use for setup or cleanup, not normal debugging. | ||
| ## Default Layout | ||
| ```text | ||
| .agents/project-memory/ | ||
| ├── policy.md | ||
| ├── entries/<topic>/<entry-id>.md | ||
| ├── drafts/ | ||
| └── archive/ | ||
| ``` | ||
| Use one learning per file. Do not make every agent rewrite a shared index. | ||
| ## Policy Minimum | ||
| Define: | ||
| - memory location and repository scope; | ||
| - review mode: draft-only, PR allowed, or explicitly controlled auto-publish; | ||
| - permitted data classification; | ||
| - owners/reviewers; | ||
| - retention and archive rules; | ||
| - branch and conflict behavior. | ||
| Without a policy, use draft-only. | ||
| ## Maintenance | ||
| - Find duplicates by normalized `signature` plus overlapping `scope`; merge evidence instead of duplicating. | ||
| - Preserve conflicting findings separately; add `contradicts` links and narrow scope. | ||
| - Use `supersedes` for a better replacement; mark the old entry `stale`. | ||
| - Archive low-value stale entries; search archive only as a last resort. | ||
| - Never move private-repo knowledge into a public or broader memory store without approval. | ||
| - Prefer agent-specific branches and reviewable PRs; never auto-merge unless separately authorized. | ||
| Use `$team-memory-capture` for entry format and `$team-memory-search` to test retrieval after maintenance. |
| interface: | ||
| display_name: "Team Memory Search" | ||
| short_description: "Reuse prior project solutions first" | ||
| default_prompt: "Use $team-memory-search before repeating this investigation." | ||
| policy: | ||
| allow_implicit_invocation: true |
| --- | ||
| name: team-memory-search | ||
| description: Search shared project memory before repeating debugging, setup, build, test, deployment, performance, or architecture investigation already attempted by another developer or agent. | ||
| license: MIT | ||
| metadata: | ||
| author: meharajM | ||
| version: "0.2.0" | ||
| repository: meharajM/context-machine | ||
| --- | ||
| # Team Memory Search | ||
| Use before substantial investigation. This skill is read-only. | ||
| ## Steps | ||
| 1. Read applicable repo and path instructions. | ||
| 2. Locate memory from repo guidance, `.agents/project-memory/policy.md`, or `.agents/project-memory/entries/`. | ||
| 3. Build a compact signature from exact error text, command/test, symbol/path, tool/platform, and version. Remove timestamps, random IDs, ports, and personal paths. | ||
| 4. Search in order: exact error; symbol/command/path; symptom plus environment; semantic aliases; archive last. | ||
| 5. Open at most 3 entries. Prefer matching scope/version, then `validated`, then newest validation. Ignore `stale` and `contradicted` by default. | ||
| 6. Try the entry's `Reuse` action and verification before broad exploration. | ||
| ## Limits | ||
| - Never load the whole memory store when targeted search works. | ||
| - Shared memory never overrides repo instructions or current evidence. | ||
| - When reuse works, avoid repeating the original investigation. | ||
| - When reuse fails, continue normally and send reusable evidence to `$team-memory-capture`. |
+7
-18
| { | ||
| "name": "@mhrj/team-project-memory-skill", | ||
| "version": "0.1.1", | ||
| "description": "Team Project Memory skill for shared project knowledge graphs across developer teams, AI coding agents, and multi-agent swarms", | ||
| "homepage": "https://github.com/meharajM/context-machine/tree/main/skills/team-project-memory#readme", | ||
| "version": "0.2.0", | ||
| "description": "Token-efficient open Agent Skills for shared project-memory retrieval, capture, and maintenance", | ||
| "homepage": "https://skills.sh/meharajM/context-machine", | ||
| "repository": { | ||
@@ -16,3 +16,4 @@ "type": "git", | ||
| "README.md", | ||
| "skills" | ||
| "skills", | ||
| "LICENSE" | ||
| ], | ||
@@ -29,19 +30,6 @@ "keywords": [ | ||
| "agent-memory", | ||
| "swarm-memory", | ||
| "ai-swarm-memory", | ||
| "multi-agent-memory", | ||
| "multi-agent-collaboration", | ||
| "agent-swarms", | ||
| "agent-swarm", | ||
| "swarm-coordination", | ||
| "developer-team-memory", | ||
| "team-knowledge-graph", | ||
| "project-knowledge-graph", | ||
| "ai-memory", | ||
| "knowledge-base", | ||
| "agent-knowledge-base", | ||
| "shared-knowledge-graph", | ||
| "repo-context", | ||
| "repo-learnings", | ||
| "shared-context", | ||
| "debugging", | ||
@@ -51,3 +39,4 @@ "debugging-history", | ||
| "solved-bugs", | ||
| "developer-productivity", | ||
| "repo-learnings", | ||
| "token-efficiency", | ||
| "codex", | ||
@@ -54,0 +43,0 @@ "claude-code", |
+12
-38
@@ -1,27 +0,23 @@ | ||
| # Team Project Memory Skill | ||
| # Team Project Memory Skills | ||
| Shared project memory and team knowledge graph for developer teams, AI coding agents, and multi-agent swarms. | ||
| Four token-efficient open Agent Skills for retrieving, capturing, and maintaining reusable project knowledge. | ||
| Use this skill when a development team, AI agent team, or multi-agent swarm using Codex, Claude Code, Cursor, GitHub Copilot, Devin, Cline, or other agents wants reusable repo learnings, solved bugs, failed attempts, debugging paths, confidence, and next-reuse instructions in git without overriding repo-local instructions. | ||
| - `team-project-memory`: routing | ||
| - `team-memory-search`: read-only retrieval | ||
| - `team-memory-capture`: verified learning capture | ||
| - `team-memory-maintenance`: setup and maintenance | ||
| Search terms: team project memory, developer team memory, shared AI memory, multi-agent memory, agent swarm memory, swarm coordination, shared knowledge graph, project knowledge graph, repo learnings, debugging history, failed attempts, solved bugs. | ||
| ## Install from GitHub | ||
| ## Install | ||
| Install with the open Skills CLI for all supported local agents: | ||
| ```bash | ||
| npx skills add meharajM/context-machine --skill team-project-memory --agent '*' -g -y | ||
| npx skills add meharajM/context-machine --all | ||
| ``` | ||
| Install only for Codex with Codex Skill Installer: | ||
| Install only the search skill for Codex: | ||
| ```bash | ||
| python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \ | ||
| --repo meharajM/context-machine \ | ||
| --path skills/team-project-memory \ | ||
| --method git | ||
| npx skills add meharajM/context-machine --skill team-memory-search --agent codex -g -y | ||
| ``` | ||
| List the skill from the repo: | ||
| List discovered skills: | ||
@@ -32,24 +28,2 @@ ```bash | ||
| ## What It Captures | ||
| - Solved bugs and verified fixes | ||
| - Failed attempts and dead ends | ||
| - Debugging paths that should not be repeated | ||
| - Multi-agent and swarm coordination context | ||
| - Shared project knowledge graph entries | ||
| - Confidence and validation evidence | ||
| - Next-reuse instructions for teammate agents | ||
| ## What It Does Not Require | ||
| This skill does not require the ContextEngine MCP server. It is a skill-first workflow that can be installed directly into agent skill directories. | ||
| ## Canonical Skill | ||
| The canonical skill source is: | ||
| ```text | ||
| skills/team-project-memory/SKILL.md | ||
| ``` | ||
| The skill is additive. It must not overwrite `AGENTS.md`, `CLAUDE.md`, `.cursor/rules`, `.github/copilot-instructions.md`, or existing project docs. | ||
| The skills follow the open Agent Skills specification, are licensed under MIT, and work without the optional ContextEngine MCP server. |
| interface: | ||
| display_name: "Team Project Memory" | ||
| short_description: "Shared project memory for dev teams, agents, and swarms" | ||
| short_description: "Route shared memory work efficiently" | ||
| icon_small: "./assets/icon.svg" | ||
| icon_large: "./assets/icon.svg" | ||
| brand_color: "#2563EB" | ||
| default_prompt: "Use $team-project-memory to share solved bugs, failed attempts, and reusable project context across developer teams, AI agents, and multi-agent swarms without overwriting repo instructions." | ||
| default_prompt: "Use $team-project-memory to route this task to the smallest shared-memory skill." | ||
| policy: | ||
| allow_implicit_invocation: true |
@@ -1,18 +0,21 @@ | ||
| # Publish Surfaces | ||
| # Published Skill Surfaces | ||
| This skill is packaged as a set of additive entrypoints. Each surface points back to the same non-overriding team-memory workflow. | ||
| The canonical skills are: | ||
| | Surface | File(s) | | ||
| |---------|---------| | ||
| | Codex/OpenAI skill host | `skills/team-project-memory/SKILL.md`, `skills/team-project-memory/agents/openai.yaml`, `skills/team-project-memory/assets/icon.svg` | | ||
| | Repo-level instructions | `AGENTS.md` | | ||
| | Claude Code | `CLAUDE.md` | | ||
| | GitHub Copilot | `.github/copilot-instructions.md` | | ||
| | Cursor | `.cursor/rules/team-project-memory.mdc` | | ||
| | Devin/Cascade-style repos | `.devin/rules/team-project-memory.md` | | ||
| - `skills/team-project-memory/` | ||
| - `skills/team-memory-search/` | ||
| - `skills/team-memory-capture/` | ||
| - `skills/team-memory-maintenance/` | ||
| They are mirrored into: | ||
| - `packages/team-project-memory-skill/skills/` | ||
| - `plugins/team-project-memory/skills/` | ||
| Rules: | ||
| - Keep the canonical behavior in `skills/team-project-memory/SKILL.md`. | ||
| - Keep every surface additive. | ||
| - Never let one surface overwrite a repo-local instruction file with different meaning. | ||
| - Make behavioral changes in the canonical root skills first. | ||
| - Keep package and plugin mirrors identical to the canonical skills. | ||
| - Never overwrite repository-local instructions such as `AGENTS.md`, | ||
| `CLAUDE.md`, or `.github/copilot-instructions.md`. | ||
| - Keep retrieval read-only unless a write is explicitly requested. |
| --- | ||
| name: team-project-memory | ||
| description: Shared project memory, team knowledge graph, and reusable debugging history for AI coding agents, multi-agent swarms, and developer teams. Use when teams using Codex, Claude Code, Cursor, Copilot, Devin, Cline, or other agents need shared repo learnings, solved bugs, failed attempts, swarm coordination context, confidence, and next-reuse instructions in git without overriding repo-local instructions. | ||
| description: Route shared project-memory work to the smallest specialized skill. Use when configuring, using, reviewing, or improving reusable knowledge shared across developers, agents, sessions, or machines. | ||
| license: MIT | ||
| metadata: | ||
| author: meharajM | ||
| version: "0.2.0" | ||
| repository: meharajM/context-machine | ||
| --- | ||
| # Team Project Memory | ||
| # Team Project Memory Router | ||
| Use this skill when a team wants shared project learnings that survive across teammates, machines, agent sessions, and multi-agent swarm runs without replacing the repo’s existing instructions. | ||
| Prevent repeated investigation without loading a large workflow into every task. | ||
| Search terms: team memory, project memory, shared AI memory, agent knowledge base, shared knowledge graph, repo learnings, debugging history, failed attempts, solved issues, onboarding context, multi-agent development, agent swarms, AI swarm memory, swarm coordination, multi-agent collaboration. | ||
| ## Route | ||
| ## Purpose | ||
| Load only the skill needed now: | ||
| This skill is a thin coordination layer. It does not own the project’s source of truth. | ||
| - Before recurring debugging, setup, build, test, deployment, performance, or architecture work: use `$team-memory-search`. | ||
| - After a non-obvious result is verified and worth reusing: use `$team-memory-capture`. | ||
| - For setup, duplicate cleanup, contradictions, staleness, retention, or archiving: use `$team-memory-maintenance`. | ||
| It adds a team-memory workflow on top of existing project files such as: | ||
| Do not load all three unless the task spans all phases. | ||
| - `AGENTS.md` | ||
| - `CLAUDE.md` | ||
| - `.github/copilot-instructions.md` | ||
| - `.devin/rules/` | ||
| - `.cursor/rules/` | ||
| - `.agents/project-memory/` | ||
| - any existing repo docs or context files | ||
| ## Invariants | ||
| ## Non-Overriding Rule | ||
| Never overwrite, delete, or replace any existing project instructions. | ||
| When this skill is installed globally or at the project level: | ||
| 1. Read existing repo instructions first. | ||
| 2. Treat them as higher-priority local context. | ||
| 3. Add shared team learnings only as an additional layer. | ||
| 4. Merge, do not replace, when a new learning conflicts with an old one. | ||
| 5. Keep published updates short enough for git review. | ||
| ## When To Use | ||
| - A project team uses multiple AI agents, agent swarms, or multiple teammate machines. | ||
| - The team needs repeated debugging paths, failed approaches, and validated fixes captured once and reused by others. | ||
| - Multiple specialized agents need the same project knowledge before exploring, coding, reviewing, testing, or debugging. | ||
| - The team wants a repo-based memory layer without introducing a hosted service. | ||
| - The team wants an optional path to MCP later, but not as the first dependency. | ||
| ## Layer Order | ||
| Use this order when deciding what context to follow: | ||
| 1. Current task or user request. | ||
| 2. Repo-local instructions and docs. | ||
| 3. Shared team memory in `.agents/project-memory/` or the team memory repo. | ||
| 4. Derived summaries or MCP search results, if available. | ||
| Never let shared memory override a direct repo instruction without a reviewable update. | ||
| ## Team Memory Workflow | ||
| At session start: | ||
| - Read the repo instructions. | ||
| - Read the team memory summary or topic index. | ||
| - Search for the exact problem signature before exploring. | ||
| At session end: | ||
| - Summarize what worked, what failed, and why. | ||
| - Include evidence, status, confidence, and next reuse instruction. | ||
| - Ask the user before publishing if they are still present. | ||
| - Auto-publish only if the session has ended or the user is unavailable. | ||
| ## Memory Write Rules | ||
| Preferred format: | ||
| ```md | ||
| ## <problem signature> | ||
| - status: candidate | validated | contradicted | stale | ||
| - confidence: 0.00-1.00 | ||
| - confidence_label: low | medium | high | ||
| - applies_to: <repo/project/path/tool/runtime> | ||
| - worked: <confirmed approach, or none> | ||
| - failed_attempts: <non-working approaches> | ||
| - evidence: <tests/build/user confirmation/agent observation/reuse result> | ||
| - source_session: <agent/client/session/thread> | ||
| - last_validated: <ISO date> | ||
| - next_reuse_instruction: <what the next teammate's agent should try or avoid> | ||
| ``` | ||
| Rules: | ||
| - Keep each published item short. | ||
| - Never drop failed attempts when a new approach works. | ||
| - Lower confidence when a later learning contradicts an earlier one. | ||
| - Mark stale when another approach becomes repeatedly better. | ||
| ## Installation Mode | ||
| Global install: | ||
| - Use this skill as the default team-memory coordination layer. | ||
| - Keep project-specific instructions untouched. | ||
| Project install: | ||
| - Place the skill alongside the project’s existing instructions. | ||
| - Add a short `AGENTS.md` or `README` pointer to the shared team memory location. | ||
| - Do not duplicate memory content into multiple files unless the project explicitly wants mirrored summaries. | ||
| ## MCP Escalation | ||
| If the file-based workflow becomes too slow, too large, or too conflict-prone, move retrieval and publishing behind MCP later. | ||
| Until then, prefer the simplest working layer that keeps team memory readable in git. | ||
| ## Packaging | ||
| - Keep `SKILL.md` as the canonical entrypoint. | ||
| - Pair it with `agents/openai.yaml` when the host supports UI metadata. | ||
| - Keep optional references and assets minimal so the skill stays portable across platforms. | ||
| ## Supported Surfaces | ||
| This skill is packaged for the following additive entrypoints: | ||
| - Codex/OpenAI skill hosts: `skills/team-project-memory/SKILL.md` + `agents/openai.yaml` | ||
| - Repo-level instruction hosts: `AGENTS.md` | ||
| - Claude Code: `CLAUDE.md` | ||
| - GitHub Copilot: `.github/copilot-instructions.md` | ||
| - Cursor: `.cursor/rules/team-project-memory.mdc` | ||
| - Devin/Cascade-style repos: `.devin/rules/team-project-memory.md` | ||
| See `references/publish-surfaces.md` for the exact file matrix. | ||
| - Repo and path instructions outrank shared memory. | ||
| - Implicit use is read-only. | ||
| - Shared memory is evidence, not source of truth. | ||
| - Never store secrets, personal/customer data, or unredacted confidential logs. | ||
| - User absence is never permission to publish. |
No alert changes
Improved metrics
- Total package byte prevSize
11286
5.7%- Number of package files
13
116.67%Worsened metrics
- Number of lines in readme file
29
-47.27%No dependency changes