agentleFS
Cody-SimsCody-Sims/agent-skills/skills/shadow-architecture/SKILL.md

shadow-architecture

Skill0 starsChanged 4 months agoMITView on GitHub

Maintains repository-owned .shadow graphs and approved decision history. Use when creating or migrating .shadow, recording approved decisions, maintaining indexes or schemas, superseding accepted history, or rendering reviewed records; not for observation discovery, future-state proposals, or drift checks.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read authority: Read workspace instructions, then .shadow/README.md and its declared index/schema.

    line 30
  2. Distinctive:

    Setup/migrate: If absent, create only a README, declared index/schema, and decision location.

    line 37
  3. Distinctive:

    Update the schema-compatible record and authoritative index together.

    line 54
  4. Very specific:

    Supersede: Never overwrite accepted substance with a new choice.

    line 55
  5. Distinctive:

    Render: Use authoritative reviewed records/indexes.

    line 61
  6. Very specific:

    Stop for human approval before acceptance, changing accepted meaning, supersession, incompatible README/index/schema changes, or converting …

    line 70
  7. Distinctive:

    Run the declared validator, or parse declared machine-readable files.

    line 80
  8. Distinctive:

    Check unique IDs, supported kinds/statuses, resolvable typed relations and required inverse links, approval/provenance, index agreement, …

    line 81

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Shadow Architecture

Maintain durable, reviewable .shadow/ memory. Repository instructions, .shadow/README.md, and its declared index/schema are authoritative, not a fixed reference layout. Read the Shadow v2 reference contract before setup, migration, or contract changes.

Evidence Rules

Separate verified evidence, inference, and unknowns; cite repository-relative code/tests/docs or stable issue/PR/commit references for material claims. Never invent intent or rationale; use unknown. Candidates, proposals, and drift reports are review inputs, not accepted decisions. Approval may produce a decision but never retroactively changes the source artifact's record kind. Rendered output is disposable, not authority.

Workflow

  1. Read authority. Read workspace instructions, then .shadow/README.md and its declared index/schema. Inventory kinds, statuses, IDs, relations, approvals, provenance, and derived output. Preserve valid local fields, paths, formats, and older records absent an approved migration.
  2. Bound the operation. Select setup, migrate, maintain, record, supersede, or render; execute only applicable steps below. Do not broaden maintenance into observation, ideation, or drift analysis.
  3. Setup/migrate. If absent, create only a README, declared index/schema, and decision location. Document authoritative, awaiting-review, and derived files; stable IDs, kinds, lifecycle/relation types, evidence/approval fields, and run provenance. Preserve existing readable shapes. Prefer aliases, optional fields, schema unions, or version markers before destructive rewrites. Record migration provenance and compatibility gaps. Obtain human approval before changing the authoritative contract.
  4. Maintain. Schema must accept every README-promised record; synchronize index paths/statuses with authoritative records, keep IDs stable and targets resolvable. Preserve unknowns, run/source provenance, and approvals through format changes. Keep derived output regenerable and non-authoritative. Report incompatible legacy records; never silently drop fields.
  5. Record. Identify reviewed input/evidence; assign a stable ID and present-tense statement. Store verified rationale or unknown, evidence, inference, unknowns, anchors, relations, source provenance, and approval state. observed means evidence-supported current architecture, not accepted; proposed means not yet adopted; accepted requires recorded human approval. Update the schema-compatible record and authoritative index together.
  6. Supersede. Never overwrite accepted substance with a new choice. Create a new approved decision with its own ID, evidence, rationale, and provenance; link it to the old record with supersedes. Mark the old record superseded and link its replacement, preserving statement/rationale/evidence/approval/ provenance. Atomically update the index so both records and links remain visible.
  7. Render. Use authoritative reviewed records/indexes. Exclude unapproved candidates unless the requested view labels them explicitly. Write only under the declared derived location; identify generator/command, source IDs, available source revisions/hashes, and generation time. Label disposable derived output. Never alter decision text, status, approval, evidence, or rationale while rendering.

Approval Gates

Stop for human approval before acceptance, changing accepted meaning, supersession, incompatible README/index/schema changes, or converting any candidate/proposal/drift report into a decision. Bind approval to the exact reviewed revision; changed substance needs new approval. Passing validation, generated text, and silence are not approval. Mechanical index synchronization or reviewed-output regeneration needs no new architecture approval only when repository policy permits.

Validation

  1. Run the declared validator, or parse declared machine-readable files.
  2. Check unique IDs, supported kinds/statuses, resolvable typed relations and required inverse links, approval/provenance, index agreement, declared evidence/anchor conventions, and explicit missing rationale.
  3. Confirm legacy readability and derived source identity/non-authoritative indexing. Review the diff for invented rationale, history rewrites, or accidental observation, dreaming, or drift work.

Response

Report operation, authoritative changes, lifecycle transitions, approval used/ needed, preserved compatibility, rendered records, validation/results, unknowns, and handoffs. Use shadow-observe for candidate discovery, shadow-dream for future states, shadow-drift for read-only comparison, codebase-exploration for orientation, and documentation-maintenance for ordinary docs.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.