Support markdown steering

.github/workflows/advisory-uk-ai-operational-resilience.md

@@ -286,6 +286,8 @@ steps:
         core.info(`Wrote bounded advisory evidence for ${payload.repository}.`);
 ---
 
+      {{#runtime-import? .github/aw/advisory.md}}
+
 <!-- Advisory outputs are advisory and non-binding. This workflow provides no guarantee of completeness, correctness, accuracy, or alignment with current UK government AI open-code and vulnerability-risk guidance. -->
 
 # Advisory / UK AI Operational Resilience

.github/workflows/advisory.md

@@ -105,6 +105,8 @@ safe-outputs:
     max: 50
 ---
 
+{{#runtime-import? .github/aw/advisory.md}}
+
 <!-- Advisory outputs are advisory and non-binding. This workflow provides no guarantee of completeness, correctness, accuracy, or alignment with current UK government AI open-code and vulnerability-risk guidance. -->
 
 # Advisory

.github/workflows/ambient-context-agents-md-curator.md

@@ -457,6 +457,8 @@ steps:
         core.info(`Ambient context evidence written to ${OUT}`);
 ---
 
+{{#runtime-import? .github/aw/ambient-context.md}}
+
 You are the AGENTS.md Curator. You maintain the ambient context of one repository: the instructions every agent session reads before doing anything else. You never edit the repository yourself. You publish one issue containing the evidence and a ready-to-run agentic prompt that a coding agent or maintainer can execute to apply a small, verifiable `AGENTS.md` diff.
 
 ## Inputs

.github/workflows/ambient-context-skills-curator.md

@@ -285,6 +285,8 @@ steps:
         core.info(`Skill evidence written to ${OUT}`);
 ---
 
+{{#runtime-import? .github/aw/ambient-context.md}}
+
 You are the Skills Curator. You keep one repository's agent skills useful and cheap: procedures live in skills, facts live in `AGENTS.md`, and every skill earns its place. You never edit the repository yourself. You publish one issue containing the evidence and a ready-to-run agentic prompt.
 
 ## Inputs

.github/workflows/ambient-context.md

@@ -98,6 +98,8 @@ safe-outputs:
     max: 20
 ---
 
+{{#runtime-import? .github/aw/ambient-context.md}}
+
 # Ambient Context
 
 Package orchestrator for the ambient context that agents read before they do anything else: `AGENTS.md` and the repository's agent skills. Ambient context decays silently, so this package runs on a weekly-or-slower cadence, selects repositories whose instructions have drifted the most from repository reality, and dispatches curators that propose evidence-backed pruning and refresh work.

.github/workflows/aw-failures-investigator.md

@@ -346,6 +346,8 @@ steps:
         core.info(`Existing tracking issues: ${existingTrackingIssues.length}`);
 ---
 
+      {{#runtime-import? .github/aw/aw-failures.md}}
+
 You are the AW Failure Investigator — a worker that analyzes recent GitHub Agentic Workflow failures in one target repository, buckets them into failure clusters, and files focused fix issues for the buckets that are not already tracked.
 
 ## Workspace Layout

.github/workflows/aw-failures.md

@@ -97,6 +97,8 @@ safe-outputs:
     max: 50
 ---
 
+{{#runtime-import? .github/aw/aw-failures.md}}
+
 # AW Failures
 
 Package orchestrator for organization-wide agentic workflow failure triage. Use the shared control plane to select repositories that run their own custom Agentic Workflows, then dispatch `aw-failures-investigator` once per selected repository. The orchestrator only selects and ranks repositories; the worker owns failure bucketization and issue creation inside each target repository.

.github/workflows/aw-maintenance-upgrade.md

@@ -256,6 +256,8 @@ steps:
       key: aw-maintenance-gh-aw-releases-${{ github.run_id }}
 ---
 
+{{#runtime-import? .github/aw/aw-maintenance.md}}
+
 You are the AW Maintenance / Upgrade worker — you keep one target repository's GitHub Agentic Workflows (gh-aw) current by detecting available releases, running `gh aw upgrade` to compute the upgrade diff, and filing one issue a maintainer can assign to Copilot to open the upgrade pull request. Traditional, hand-written GitHub Actions YAML is out of scope for this worker.
 
 ## Workspace Layout

.github/workflows/aw-maintenance.md

@@ -97,6 +97,8 @@ safe-outputs:
     max: 50
 ---
 
+{{#runtime-import? .github/aw/aw-maintenance.md}}
+
 # AW Maintenance
 
 Package orchestrator for organization-wide GitHub Agentic Workflows (gh-aw) maintenance. Use the shared control plane to select repositories that install their own GitHub Agentic Workflows, then dispatch `aw-maintenance-upgrade` once per selected repository. The orchestrator only selects and ranks repositories; the worker owns release detection, the `gh aw upgrade` run, and issue filing inside each target repository.

.github/workflows/dependabot-release-train-updater.md

@@ -385,6 +385,8 @@ timeout-minutes: 60
 source: githubnext/central-agentic-ops/.github/workflows/dependabot-release-train-updater.md@main
 ---
 
+{{#runtime-import? .github/aw/dependabot.md}}
+
 You are a dependency reliability and supply-chain maintenance agent for the checked-out safe-output repository.
 Your job is to turn dependency maintenance into one safe, reviewable, manifest-aware outcome.
 You do **not** auto-merge. You create pull requests, PR updates, comments, issues, or noop results through safe outputs only.

.github/workflows/dependabot.md

@@ -102,6 +102,8 @@ safe-outputs:
 source: githubnext/central-agentic-ops@2de9130ff1709fccdacbe5261fd5da71995e6721
 ---
 
+{{#runtime-import? .github/aw/dependabot.md}}
+
 # Dependabot
 
 Package orchestrator for organization-wide dependency release-train maintenance. Use the shared control plane to select target repositories and dispatch `dependabot-release-train-updater`; keep dispatch repository-scoped and let the updater own manifest-aware bundle construction inside each selected repository.

.github/workflows/eu-cra-compliance-article-14-reporting-readiness.md

@@ -111,6 +111,8 @@ safe-outputs:
 timeout-minutes: 30
 ---
 
+{{#runtime-import? .github/aw/eu-cra-compliance.md}}
+
 # EU CRA Advisor / Article 14 Reporting Readiness
 
 Audit operational readiness for Article 14 reporting. This worker never decides reportability without human review and never sends a notification.

.github/workflows/eu-cra-compliance-conformity-release-evidence.md

@@ -111,6 +111,8 @@ safe-outputs:
 timeout-minutes: 30
 ---
 
+{{#runtime-import? .github/aw/eu-cra-compliance.md}}
+
 # EU CRA Advisor / Conformity Release Evidence
 
 Audit evidence used by human conformity and market-release gates. Do not select a conformity route, approve a declaration, authorize CE marking, or approve market release.

.github/workflows/eu-cra-compliance-scope-classifier.md

@@ -109,6 +109,8 @@ safe-outputs:
 timeout-minutes: 25
 ---
 
+{{#runtime-import? .github/aw/eu-cra-compliance.md}}
+
 # EU CRA Advisor / Scope Classifier
 
 Build a repository-level CRA scope evidence record. Assist human decision-makers; do not issue a legal conclusion.

.github/workflows/eu-cra-compliance-security-requirements-auditor.md

@@ -111,6 +111,8 @@ safe-outputs:
 timeout-minutes: 30
 ---
 
+{{#runtime-import? .github/aw/eu-cra-compliance.md}}
+
 # EU CRA Advisor / Security Requirements Auditor
 
 Audit repository evidence for applicable CRA product cybersecurity requirements. This is implementation assistance and evidence gathering, not a legal conformity decision.

.github/workflows/eu-cra-compliance-supply-chain-sbom-auditor.md

@@ -112,6 +112,8 @@ safe-outputs:
 timeout-minutes: 30
 ---
 
+{{#runtime-import? .github/aw/eu-cra-compliance.md}}
+
 # EU CRA Advisor / Supply Chain SBOM Auditor
 
 Audit repository-level software supply-chain and SBOM evidence relevant to the CRA. Do not make a legal conformity determination.

.github/workflows/eu-cra-compliance-vulnerability-handling-auditor.md

@@ -111,6 +111,8 @@ safe-outputs:
 timeout-minutes: 30
 ---
 
+{{#runtime-import? .github/aw/eu-cra-compliance.md}}
+
 # EU CRA Advisor / Vulnerability Handling Auditor
 
 Audit vulnerability-handling evidence relevant to CRA implementation. Do not decide legal conformity or submit notifications.

.github/workflows/eu-cra-compliance.md

@@ -111,6 +111,8 @@ safe-outputs:
     max: 48
 ---
 
+{{#runtime-import? .github/aw/eu-cra-compliance.md}}
+
 <!-- EU CRA Advisor is advisory and non-binding. This workflow provides no guarantee of completeness, correctness, accuracy, or alignment with the EU Cyber Resilience Act. -->
 
 # EU CRA Advisor

.github/workflows/optimization-ai-credit-auditor.md

@@ -241,6 +241,8 @@ steps:
 source: githubnext/central-agentic-ops/.github/workflows/optimization-ai-credit-auditor.md@main
 ---
 
+{{#runtime-import? .github/aw/optimization.md}}
+
 You are the Agentic Workflow Auditor — a workflow that tracks daily AI Credit (AIC) spend and token consumption, forecasts weekly and monthly cost, and maintains a historical record for trend analysis across all agentic workflows in the target repository.
 
 ## Workspace Layout

.github/workflows/optimization-ai-credit-optimizer.md

@@ -257,6 +257,8 @@ steps:
 source: githubnext/central-agentic-ops/.github/workflows/optimization-ai-credit-optimizer.md@main
 ---
 
+{{#runtime-import? .github/aw/optimization.md}}
+
 You are the Agentic Workflow Optimizer. Pick one high AI credit workflow, audit recent runs, and create a conservative optimization issue with measurable improvements. Your recommendations may include prompt, tool, reliability, setup-prefix, and inline sub-agent improvements when the evidence supports them.
 
 ## Objectives

.github/workflows/optimization.md

@@ -98,6 +98,8 @@ safe-outputs:
 source: githubnext/central-agentic-ops@2de9130ff1709fccdacbe5261fd5da71995e6721
 ---
 
+{{#runtime-import? .github/aw/optimization.md}}
+
 # Optimization
 
 ## Discovery

docs/configuration.md

@@ -37,6 +37,14 @@ gh variable set CENTRAL_AGENTIC_OPS_DEPENDABOT_MAX_REPOS \
 
 Add an App or PAT when the target is private or internal. Keep the mode at `review` until the promotion checks pass.
 
+## Markdown Steering
+
+Each operation orchestrator can load repository-specific instructions from `.github/aw/<operation>.md` in the control repository. For example, `.github/aw/dependabot.md` can describe organization-specific dependency priorities, repositories to prefer or avoid, or additional evidence to consider.
+
+The supported operation names are `advisory`, `ambient-context`, `aw-failures`, `aw-maintenance`, `dependabot`, `eu-cra-compliance`, and `optimization`. These files are optional runtime imports: an operation continues with its packaged instructions when its steering file does not exist. Because steering files are separate from package-owned workflow sources, `gh aw update` does not overwrite them.
+
+Keep steering instructions within the operation's existing permissions, safety policy, and dispatch limits. Steering can refine selection and prioritization, but it cannot grant tools, credentials, permissions, or safe-output capabilities.
+
 ## Repository Variables
 
 | Name | Scope | Required | Default | Purpose |

tests/unit/workflow-contract.test.mjs

@@ -714,6 +714,37 @@ test("orchestrators expose scheduled variables and independent manual inputs", (
   }
 });
 
+test("operation workflows optionally load per-operation markdown steering", () => {
+  for (const [name, operation] of [
+    ["advisory.md", "advisory"],
+    ["advisory-uk-ai-operational-resilience.md", "advisory"],
+    ["ambient-context.md", "ambient-context"],
+    ["ambient-context-agents-md-curator.md", "ambient-context"],
+    ["ambient-context-skills-curator.md", "ambient-context"],
+    ["aw-failures.md", "aw-failures"],
+    ["aw-failures-investigator.md", "aw-failures"],
+    ["aw-maintenance.md", "aw-maintenance"],
+    ["aw-maintenance-upgrade.md", "aw-maintenance"],
+    ["dependabot.md", "dependabot"],
+    ["dependabot-release-train-updater.md", "dependabot"],
+    ["eu-cra-compliance.md", "eu-cra-compliance"],
+    ["eu-cra-compliance-article-14-reporting-readiness.md", "eu-cra-compliance"],
+    ["eu-cra-compliance-conformity-release-evidence.md", "eu-cra-compliance"],
+    ["eu-cra-compliance-scope-classifier.md", "eu-cra-compliance"],
+    ["eu-cra-compliance-security-requirements-auditor.md", "eu-cra-compliance"],
+    ["eu-cra-compliance-supply-chain-sbom-auditor.md", "eu-cra-compliance"],
+    ["eu-cra-compliance-vulnerability-handling-auditor.md", "eu-cra-compliance"],
+    ["optimization.md", "optimization"],
+    ["optimization-ai-credit-auditor.md", "optimization"],
+    ["optimization-ai-credit-optimizer.md", "optimization"],
+  ]) {
+    assert.match(
+      workflow(name),
+      new RegExp(`\\{\\{#runtime-import\\? \\.github/aw/${operation}\\.md\\}\\}`),
+    );
+  }
+});
+
 test("review destinations must be isolated, accessible, and private", () => {
   const precompute = workflow("shared/control-precompute.md");