Audit Repository
Repository continuity composition
For repository-scoped work, compose maintain-repository-continuity in
Resume mode before selecting work. This skill is continuity read-only: do
not create, refresh, or otherwise mutate CONTINUITY.md unless a separate
authorized repository-changing workflow takes ownership of that handoff.
- Contribute: Audit scope, observed findings, confidence, evidence gaps, and executed checks
- Never claim: Permission to modify CONTINUITY.md, apply fixes, or convert recommendations into completed work
Purpose
Execute one reusable procedure governed by approved auditor 2.0.0. The draft
repository-audit-data 1.0.0 extension supplies composable profile/data
interfaces and depends on that approved foundation; it does not change its
human lifecycle. This skill remains draft pending deterministic and host
qualification under Aether #86.
Primary question:
What does observable repository evidence show about quality, alignment, risk, and strengths?
Required Inputs
Resolve:
- governing specification and version
- current source or repository state
- scope and constraints
- upstream architecture or evidence
- output location
- validation expectations
- unresolved decisions
- explicit profile IDs/versions or a named composite from the exact reviewed
profile registry
- for organization scope, an explicit authorized repository/revision roster
and the applicable owner public-admission evidence
Missing evidence must remain visible.
Workflow
- Resolve request, recorded defaults, scope and read-only constraints. Select
one profile, several explicit profiles, or
holistic; use the exact registry
resolver and retain both requested and resolved ID/version references.
Experimental opportunities require explicit additional selection.
- Authenticate the supplied roster through its owner admission boundary before
reading excluded/private findings, identities, counts or hashes. Do not
discover an organization or silently expand the selected repository scope.
- Inspect repository context in the approved discovery order, then existing
audits. Record exact represented revisions, included/excluded/inspected/
uninspected scope, tools and actual limitations. Missing context remains
unknown, blocked or unsupported rather than guessed applicability.
- Gather evidence within scope. Read-only diagnostics require actual host and
user authorization; a read/search-only agent can consume supplied captured
evidence or hand off requested diagnostics, not acquire execute authority.
- Follow each leaf profile's focus, required evidence and exclusions using
this same procedure. Label every finding with its original leaf ID/version
and evidence classification; overlapping results do not erase origins.
- Assign the approved classification, severity, confidence, status, effort and
impact. Keep defects, policy observations, opportunities, experiments and
strengths distinct. An absent feature or speculative recommendation is not
a confirmed failure. State missing evidence for low-confidence claims.
- Preserve explicit profile/run completion, uninspected scope, continuation,
stale evidence, uncertain history and known prior-audit links. No missing
record, later capture or changed title resolves an older finding.
- Propose bounded validation and candidate issues as suggestions. Do not
create issues, apply remediation, accept risk, or ratify architecture.
- Return the canonical report and sidecar candidates to an explicitly
authorized output writer. If this execution has new-report write authority,
write only new unique report/sidecar files, never source/history. Bind actual
Markdown bytes and validate the same authored model through the exact
machine interface. Record every unrun
validation and downstream qualification gap.
Output Contract
Primary output:
audits/{audit-name}-{utc-timestamp}.md
The typed sidecar uses the same filename stem with .json, exact machine
schema/profile versions, canonical Markdown path/hash and source provenance.
It is a projection from the report's authored finding model, not a separately
invented assessment or prose scraper. Record assumptions, evidence gaps,
validation state, unresolved questions, positive observations or their absence,
and separately authorized downstream actions.
Constraints
- Follow the governing specification.
- Preserve provenance and uncertainty.
- Do not invent authority, evidence, or current behavior.
- Do not silently expand scope.
- Do not claim completion when required validation is missing.
- Keep proposed downstream work separate from authorized execution.
- Complete scope is not pass, conformance, deployment or human acceptance.
- Preserve original profile/evidence classification and independent finding
history; experiments remain informational and recommendation-only.
- Exclude credentials, private conversation, raw logs and machine-local paths.
Completion Criteria
Staged reconciliation and remaining maturity
A staged candidate at .staging/skills/repository-audit/ covers similar
ground under the name repository-audit. The canonical skill is named
audit-repository and is governed by the auditor specification.
The source reconciliation records the
meaningful staged/canonical delta under Aether #86/#68. It retains observed
diagnostics, current prior-audit checks, strengths, bounded backlog and partial
reporting while rejecting monolith paths and staged agent edit/execute tools.
Historical staged files remain migration evidence; none is promoted or deleted.
Single/multiple/holistic/experimental profile fixtures, malformed/stale/partial/
blocked/not-applicable cases, dependency closure and actual host/discovery
qualification remain unexecuted. Source reconciliation or a version bump does
not promote this draft skill.
1---2name: audit-repository3description: Audits an explicitly scoped repository or public organization roster through composable profiles, preserving evidence, uncertainty, strengths and suggestion-only follow-up in immutable reports and typed sidecars.4license: MIT5metadata:6 source: https://skillmd.com/skills/egohygiene/audit-repository7 aether-version: "1.2.0"8 aether-status: "draft"9 aether-spec-id: "repository-audit-data"10 aether-scope: "organization"11 aether-domain: "quality"12 aether-owners: "egohygiene"13 aether-created: "2026-08-02"14 aether-updated: "2026-11-09"15 aether-distribution-resources:16 - source: "catalog/schemas/aether.audit-profile-registry.v1.schema.json"17 destination: "references/aether.audit-profile-registry.v1.schema.json"18 - source: "catalog/schemas/aether.repository-audit-report.v1.schema.json"19 destination: "references/aether.repository-audit-report.v1.schema.json"20---2122# Audit Repository2324<!-- aether-continuity-disposition: read-only -->2526## Repository continuity composition2728For repository-scoped work, compose `maintain-repository-continuity` in29**Resume** mode before selecting work. This skill is continuity read-only: do30not create, refresh, or otherwise mutate `CONTINUITY.md` unless a separate31authorized repository-changing workflow takes ownership of that handoff.3233- **Contribute:** Audit scope, observed findings, confidence, evidence gaps, and executed checks34- **Never claim:** Permission to modify CONTINUITY.md, apply fixes, or convert recommendations into completed work3536## Purpose3738Execute one reusable procedure governed by approved `auditor` 2.0.0. The draft39`repository-audit-data` 1.0.0 extension supplies composable profile/data40interfaces and depends on that approved foundation; it does not change its41human lifecycle. This skill remains **draft** pending deterministic and host42qualification under [Aether #86](https://github.com/egohygiene/aether/issues/86).4344Primary question:4546> What does observable repository evidence show about quality, alignment, risk, and strengths?4748## Required Inputs4950Resolve:5152- governing specification and version53- current source or repository state54- scope and constraints55- upstream architecture or evidence56- output location57- validation expectations58- unresolved decisions59- explicit profile IDs/versions or a named composite from the exact reviewed60 [profile registry](references/audit-profiles.v1.json)61- for organization scope, an explicit authorized repository/revision roster62 and the applicable owner public-admission evidence6364Missing evidence must remain visible.6566## Workflow67681. Resolve request, recorded defaults, scope and read-only constraints. Select69 one profile, several explicit profiles, or `holistic`; use the exact registry70 resolver and retain both requested and resolved ID/version references.71 Experimental opportunities require explicit additional selection.722. Authenticate the supplied roster through its owner admission boundary before73 reading excluded/private findings, identities, counts or hashes. Do not74 discover an organization or silently expand the selected repository scope.753. Inspect repository context in the approved discovery order, then existing76 audits. Record exact represented revisions, included/excluded/inspected/77 uninspected scope, tools and actual limitations. Missing context remains78 unknown, blocked or unsupported rather than guessed applicability.794. Gather evidence within scope. Read-only diagnostics require actual host and80 user authorization; a read/search-only agent can consume supplied captured81 evidence or hand off requested diagnostics, not acquire execute authority.825. Follow each leaf profile's focus, required evidence and exclusions using83 this same procedure. Label every finding with its original leaf ID/version84 and evidence classification; overlapping results do not erase origins.856. Assign the approved classification, severity, confidence, status, effort and86 impact. Keep defects, policy observations, opportunities, experiments and87 strengths distinct. An absent feature or speculative recommendation is not88 a confirmed failure. State missing evidence for low-confidence claims.897. Preserve explicit profile/run completion, uninspected scope, continuation,90 stale evidence, uncertain history and known prior-audit links. No missing91 record, later capture or changed title resolves an older finding.928. Propose bounded validation and candidate issues as suggestions. Do not93 create issues, apply remediation, accept risk, or ratify architecture.949. Return the canonical report and sidecar candidates to an explicitly95 authorized output writer. If this execution has new-report write authority,96 write only new unique report/sidecar files, never source/history. Bind actual97 Markdown bytes and validate the same authored model through the exact98 [machine interface](references/audit-data-contract.md). Record every unrun99 validation and downstream qualification gap.100101## Output Contract102103Primary output:104105 audits/{audit-name}-{utc-timestamp}.md106107The typed sidecar uses the same filename stem with `.json`, exact machine108schema/profile versions, canonical Markdown path/hash and source provenance.109It is a projection from the report's authored finding model, not a separately110invented assessment or prose scraper. Record assumptions, evidence gaps,111validation state, unresolved questions, positive observations or their absence,112and separately authorized downstream actions.113114## Constraints115116- Follow the governing specification.117- Preserve provenance and uncertainty.118- Do not invent authority, evidence, or current behavior.119- Do not silently expand scope.120- Do not claim completion when required validation is missing.121- Keep proposed downstream work separate from authorized execution.122- Complete scope is not pass, conformance, deployment or human acceptance.123- Preserve original profile/evidence classification and independent finding124 history; experiments remain informational and recommendation-only.125- Exclude credentials, private conversation, raw logs and machine-local paths.126127## Completion Criteria128129- [ ] Governing specification is resolved.130- [ ] Exact requested/resolved profile versions and registry digest are retained.131- [ ] Scope and constraints are explicit.132- [ ] Required evidence was inspected.133- [ ] New outputs were written through an authorized writer or returned as134 candidates with the missing host capability explicit.135- [ ] Validation was executed or its absence documented.136- [ ] Open questions and authorization needs are visible.137138## Staged reconciliation and remaining maturity139140A staged candidate at `.staging/skills/repository-audit/` covers similar141ground under the name `repository-audit`. The canonical skill is named142`audit-repository` and is governed by the `auditor` specification.143144The [source reconciliation](references/staged-reconciliation.md) records the145meaningful staged/canonical delta under Aether #86/#68. It retains observed146diagnostics, current prior-audit checks, strengths, bounded backlog and partial147reporting while rejecting monolith paths and staged agent edit/execute tools.148Historical staged files remain migration evidence; none is promoted or deleted.149Single/multiple/holistic/experimental profile fixtures, malformed/stale/partial/150blocked/not-applicable cases, dependency closure and actual host/discovery151qualification remain unexecuted. Source reconciliation or a version bump does152not promote this draft skill.