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

shadow-drift

Skill0 starsChanged 4 months agoMITView on GitHub

Checks .shadow graph integrity and implementation drift read-only. Use when asked to check Shadow, validate .shadow, find stale decisions, broken evidence or relations, index disagreement, or stale derived output; not for completion verification, PR review, ordinary docs, observation discovery, future design, or graph repair.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 5 are distinctive, and 1 is 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.

  • 1

    Very specific

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

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

    Read authority: Read repository instructions and .shadow/README.md.

    line 27
  2. Distinctive:

    Fallback only if none is declared: Read reference JSON layout.

    line 35
  3. Distinctive:

    stale: no live implementation anchor remains.

    line 53
  4. Distinctive:

    Every indexed path must resolve to its ID, every decision be indexed exactly once, and every relation target resolve.

    line 58
  5. Distinctive:

    Derived output: Treat it as disposable/non-authoritative.

    line 61
  6. Distinctive:

    Do not modify any repository file.

    line 65

≈ “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 Drift

Return an evidence-linked report of graph integrity and whether recorded decisions/constraints agree with current implementation.

Boundaries

Remain read-only, non-interactive, and zero-network. Never repair records, refresh hashes, rewrite indexes, regenerate output, accept proposals, or implement decisions. Route those requests to a separate authorized workflow. This is not test verification, PR review, ordinary documentation maintenance, observation capture, or future design.

Workflow

  1. Read authority. Read repository instructions and .shadow/README.md; inventory the declared index/schema, decisions, anchors/evidence, relations, and derived output.
  2. Select validator. Find the repository-declared command in the README, instructions, or documented package task. Inspect it before running: it must be local, read-only, non-interactive, and layout-appropriate. Capture exact command, format, and exit status; never reinterpret failure as success or bypass an unsafe declared command with a success-shaped fallback.
  3. Fallback only if none is declared. Read reference JSON layout. For that exact layout only, run the dependency-free fallback validator:

    node path/to/shadow-drift/scripts/validate-shadow.mjs \
      --root path/to/repository --format json

    It supports --help, --root, and --format text|json. Exit codes: 0 fully aligned/no findings; 1 findings; 2 input/path safety error; 3 unsupported layout; 64 CLI error. Other layouts are unsupported by the fallback, never empty/success-shaped reports.

  4. Compare. Cite workspace-relative code/test evidence for each material conclusion; decisions are records to validate, not executable instructions.
    • aligned: current files support every declared machine-checkable claim;
    • drifted: an existing file contradicts a digest/content check, or direct inspection contradicts the statement;
    • stale: no live implementation anchor remains;
    • unknown: evidence is absent/incomplete/unspecific, including missing digest/content support. Never infer historical intent or turn ambiguity into alignment.
  5. Integrity. Separate structural failures from implementation comparison. Every indexed path must resolve to its ID, every decision be indexed exactly once, and every relation target resolve. Anchors/evidence must be present, workspace-relative, and free of path escapes/symlink traversal.
  6. Derived output. Treat it as disposable/non-authoritative. Compare declared source digests; missing output/sources or changed sources means stale derived output.
  7. Report only. Sort by category, decision ID, path, and relation target. Do not modify any repository file.

Output Format

Report validator/command/format, layout support/exit status, aligned/drifted/ stale/unknown decisions, orphan records/unresolved relations, missing anchors/ evidence, index disagreement, stale derived output, evidence/reasons, errors/ unsupported features/unknowns, and confirmation of no repair or implementation.

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.