agent-continuity-stack
AGENTS.md1 starsChanged 4 months agoMITView on GitHub
This repository publishes repository-centered continuity through an optional portable user-global operating contract, memory conventions, client adapters, lifecycle safeguards, and on-demand Agent Skills for Codex, Cursor, Claude Code, and Google Antigravity. Keep skill content independent of an individual workstation, private project, account, or repository owner. - GLOBAL_CUSTOM_INSTRUCTIONS.md is the canonical complete Expert Agent Operating Contract. Do not summarize, trim, or independently edit client copies. - instruction-pack.json records its integrity and client adapters. - manifest.json owns package inventory and versions. - Each…
- Reads credentials
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,700 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository15 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- apply-project-updatesSkill
- audit-project-updatesSkill
- design-agent-capabilitySkill
- engineer-cli-toolsSkill
- maintain-agent-instructionsSkill
- maintain-project-docsSkill
- maintain-project-memorySkill
- maintain-web-design-systemSkill
- prepare-web-assetsSkill
- red-team-technical-planSkill
- release-web-applicationSkill
- research-software-versionsSkill
- verify-software-changeSkill
- web-discovery-artifactsSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
13 rules. About 0 are common, 7 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 38
Use the portable Agent Skills core: exact SKILL.md with only name and description frontmatter.
- Distinctive: line 40
Keep activation descriptions discriminative and procedures in the body or directly linked references.
- Distinctive: line 42
Do not add an executable remote installer.
- Distinctive: line 45
Keep skill installation independent from custom-instruction installation.
- Distinctive: line 48
Keep backups, receipts, baselines, and removed copies outside every client discovery root.
- Very specific: line 50
Keep MEMORY.md local-only and Git-ignored.
- Very specific: line 57
Keep Codex AGENTS.override.md and Claude CLAUDE.local.md plus .claude/settings.local.json local-only.
- Very specific: line 62
Do not embed personal identity, home paths, emails, private hosts, credentials, device IDs, or private project schemas under skills/.
- Distinctive: line 71
Keep optional skills uninstalled by default and require collision tests before adding one to the core tier.
- Very specific: line 73
Update README.md, MEMORY.md, HANDOFF.md, ARCHITECTURE.md, CHANGELOG.md, manifests, activation cases, and client smoke tests when their …
- Distinctive: line 81
Run from the repository root:
- Very specific: line 92
Before release, run the client matrix in tests/CLIENT_SMOKE_MATRIX.md against the supported client versions.
- Distinctive: line 96
Create a release archive only from an identified clean commit.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agent Continuity Stack Instructions
Project purpose and boundary
This repository publishes repository-centered continuity through an optional portable user-global operating contract, memory conventions, client adapters, lifecycle safeguards, and on-demand Agent Skills for Codex, Cursor, Claude Code, and Google Antigravity. Keep skill content independent of an individual workstation, private project, account, or repository owner.
Authorities
GLOBAL_CUSTOM_INSTRUCTIONS.mdis the canonical complete Expert Agent Operating Contract. Do not summarize, trim, or independently edit client copies.instruction-pack.jsonrecords its integrity and client adapters.manifest.jsonowns package inventory and versions.- Each
skills/<name>/skill-manifest.jsonowns that skill's independent version. optional_skills/contains default-uninstalled capabilities declared separately in the package manifest.schemas/owns model-operated lifecycle record contracts.integrations/owns exact-tested specialist-toolkit boundaries.INSTALL_PROTOCOL.mdowns installation, update, backup, and rollback behavior.CLIENT_COMPATIBILITY.mdrecords version-sensitive discovery assumptions.ARCHITECTURE.mdrecords durable component, lifecycle, and trust boundaries.MEMORY.mdrecords verified durable project state.HANDOFF.mdrecords the concise current continuation point for another model.CLAUDE.mdis a one-line Claude Code import of the canonicalAGENTS.md; do not duplicate project instructions there.
Repository policy and current host safety instructions retain their actual client-defined precedence.
Editing rules
- Use the portable Agent Skills core: exact
SKILL.mdwith onlynameanddescriptionfrontmatter. - Keep activation descriptions discriminative and procedures in the body or directly linked references.
- Do not add an executable remote installer. Installation remains model-driven, status-first, explicitly authorized, journaled one client at a time, and best-effort recoverable.
- Keep skill installation independent from custom-instruction installation. Preserve existing instructions by default and require explicit review for guided additions or the complete contract.
- Keep backups, receipts, baselines, and removed copies outside every client discovery root.
- Keep
MEMORY.mdlocal-only and Git-ignored. Never stage, commit, release, or upload it unless the user explicitly authorizes sharing it. - For maintainer-authored commits and annotated tags, use the GitHub-linked
account
ImYourBoyRoywith@imyourboyroyas the Git name and an account-linked GitHub no-reply email. Do not use the public contact email as commit metadata, and preserve other contributors' and bot identities during any history rewrite. - Keep Codex
AGENTS.override.mdand ClaudeCLAUDE.local.mdplus.claude/settings.local.jsonlocal-only; shared policy belongs in canonicalAGENTS.mdor tracked project settings. - Reject symlinks, junctions, reparse points, and unsupported special files in copied or restored trees.
- Do not embed personal identity, home paths, emails, private hosts,
credentials, device IDs, or private project schemas under
skills/. - Maintain one workflow owner. Toolkit-managed Astro + Cloudflare implementation belongs to Portable Web Toolkit; generic skills may delegate but must not reproduce its commands or configuration systems.
- For Python runtime-manager recommendations, follow the README's Python runtime preference only when neither the user nor repository already selects a manager. Never install it or replace the project's package manager or lockfile implicitly.
- Keep optional skills uninstalled by default and require collision tests before adding one to the core tier.
- Update
README.md,MEMORY.md,HANDOFF.md,ARCHITECTURE.md,CHANGELOG.md, manifests, activation cases, and client smoke tests when their owned contract changes. - Increment a skill version for a material change to that skill. Increment the package version for a published package-level change.
Validation
Run from the repository root:
python3 scripts/validate_bundle.py
python3 scripts/inspect_contract.py source
python3 scripts/inspect_tree.py . --exclude .git --exclude .env \
--exclude MEMORY.md --exclude AGENTS.override.md \
--exclude CLAUDE.local.md --exclude .claude/settings.local.json
python3 tests/run_portable_tests.pyBefore release, run the client matrix in tests/CLIENT_SMOKE_MATRIX.md against
the supported client versions. A copied file is not evidence that a client
loaded it.
Create a release archive only from an identified clean commit. Record the commit, release date, archive name, and SHA-256 in the release notes.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

