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

shadow-observe

Skill0 starsChanged 4 months agoMITView on GitHub

Generates bounded, evidence-linked candidate architecture observations. Use when asked to run Shadow, scan current architecture, or map boundaries and dependencies for .shadow; not for unfamiliar-codebase orientation, locating change sites, decision creation or repair, future states, or drift checks.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    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:

    Do not invent rationale, intent, ownership, or decision status, create/promote decisions, propose future states, or repair drift.

    line 19
  2. Very specific:

    Read authority: Read scoped repository instructions, .shadow/README.md, and its named index/schema.

    line 29
  3. Distinctive:

    Inventory before content: Run the deterministic inventory helper only when a bounded names/structure inventory helps.

    line 36
  4. Distinctive:

    Classify: Every Verified claim needs cited direct repository evidence.

    line 42
  5. Distinctive:

    Return/write conditionally: Read the observation record guide for a requested structured report or before artifact writing.

    line 48
  6. Distinctive:

    If files were explicitly requested, re-read the declared layout/schema and write only candidate artifacts within it.

    line 51
  7. Very specific:

    Check resolvable workspace-relative anchors/evidence for verified claims, labeled cited inferences, uninvented unknowns, indexed decision …

    line 66

≈ “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 Observe

Analyze current architecture on demand, bounded and read-only. Return candidate observations, never decisions.

Boundaries

Do not invent rationale, intent, ownership, or decision status, create/promote decisions, propose future states, or repair drift. Default to a report only; candidate files require explicit user request and a declared location/schema/ unreviewed lifecycle. Use codebase-exploration for orientation/build/test/ conventions/change sites, shadow-architecture for graph/decision/index/schema/ anchor/lifecycle/render maintenance, shadow-dream for future states, and shadow-drift for read-only comparison.

Workflow

  1. Read authority. Read scoped repository instructions, .shadow/README.md, and its named index/schema. Missing declarations mean report only; never bootstrap .shadow/.
  2. Bound. State question, included/excluded paths, depth or entry limit, and stopping condition. Use the smallest slice that answers the question.
  3. Provenance. Record available source revision, relevant dirty-worktree state, commands/tools, limits, and unavailable provenance as unknown.
  4. Inventory before content. Run the deterministic inventory helper only when a bounded names/structure inventory helps. It never follows symlinks, reads contents, writes files, or uses the network. Read only necessary manifests, boundaries, public interfaces, dependency declarations, tests, and Shadow records; avoid broad scans.
  5. Classify. Every Verified claim needs cited direct repository evidence; every Inference needs a labeled bounded interpretation and cited support; every Unknown needs the missing fact and resolving evidence. Keep contradictions as observations, not automatic drift diagnoses/repairs.
  6. Relate. Link existing decision IDs only when the declared index supports them. Never manufacture IDs or convert observations into decisions.
  7. Return/write conditionally. Read the observation record guide for a requested structured report or before artifact writing. Default to the response below. If files were explicitly requested, re-read the declared layout/schema and write only candidate artifacts within it. Missing location, unreviewed lifecycle, valid fields/relations, or safe destination means no write: report the gap. Reject symlink/path escapes.

Lifecycle Guidance

Assign only Candidate: machine-generated/unreviewed. Confirmed requires the declared reviewer/process, never self-confirmation. Dismissed requires reviewer rejection/supersession/archive with reason/evidence retained where supported. Follow different declared state names without adding unsupported values or crossing these approval boundaries.

Validation

  1. Check resolvable workspace-relative anchors/evidence for verified claims, labeled cited inferences, uninvented unknowns, indexed decision IDs, scope, revision or unknown, and run provenance.
  2. For written artifacts, run the declared validator; verify schema and location. Confirm no source, decision, index, or lifecycle record changed except the exact explicitly requested, schema-required candidate artifact update.

Output Format

Return Scope (question/paths/limits/stopping reason), Run provenance (revision/worktree/tools/unavailable fields), Verified claims (anchors/evidence/related decision IDs), Inferences (support/confidence boundary), Unknowns (fact/resolving evidence), Candidate artifacts (none, or requested paths/validation), and Lifecycle next step (review to confirm/dismiss).

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.