Add staged weekly reporting by juliamuiruri4 · Pull Request #113 · Azure-Samples/JavaScript-AI-Buildathon · GitHub
Skip to content
Draft
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 19 additions & 8 deletions .github/agentic-refresh/README.md
17 changes: 17 additions & 0 deletions .github/agentic-refresh/governance.md
Original file line number Diff line number Diff line change
Expand Up @@ -96,6 +96,23 @@ With staging enabled, the compiled Phase 1 locks contain no repository write
permission. Phase 2 must deliberately remove staging and revisit permissions
before enabling real issue reporting.

## Phase 2A staged reporting guard

The weekly reporting layer remains manual-only and report-only:

- the deterministic orchestrator has `workflow_dispatch` but no schedule;
- all five Quest Master workflows accept reusable `workflow_call` inputs for an
ISO week and prior ISO week;
- the only agent-requested output is a custom job that validates and uploads a
run-scoped JSON artifact;
- the digest tolerates missing quest artifacts by recording the scope as
blocked, then creates deterministic parent and child previews as artifacts;
- no compiled or traditional workflow grants issue, branch, pull-request,
workflow-dispatch, or content write permission.

The intended `0 8 * * 1` schedule and real issue safe outputs remain Phase 2B
activation work and require default-branch trials plus the bounded token spike.

## SAML authorization

SAML authorization is needed only for local audits and human GitHub operations.
Expand Down
29 changes: 29 additions & 0 deletions .github/agentic-refresh/report-contract.md
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,25 @@
Each weekly run produces one digest covering all five quests. It creates a child
plan only for a material change.

Every quest run first emits one version 1 JSON artifact with:

- ISO week key, prior ISO week key, and exact quest slug;
- exactly one result: `material-change`, `no-material-change`, or `blocked`;
- source facts separated from repository observations and proposed changes;
- lifecycle, allowlisted evidence URL, evidence location, wording, and source
fingerprint for every official-source delta;
- learner impact, owned-path drift, asset impact, report-only CodeTour impact,
validation, risks, human decisions, and prior finding keys;
- deterministic parent key, child key, markers, source fingerprints, and finding
keys added by `scripts/weekly-refresh-report.mjs`.

A completed report must include source fingerprints. A blocked report must state
one concise reason and cannot propose changes. A no-change report cannot propose
changes. Invalid or missing quest artifacts become blocked scopes in the digest.
Source and path validation rejects adjacent URL-prefix matches, absolute paths,
backslashes, normalization changes, and traversal segments before materializing
an artifact.

A child plan must contain:

1. Quest and ISO week key.
Expand All @@ -23,3 +42,13 @@ Reports must distinguish source facts, repository observations, inferred
impact, proposed changes, and unresolved uncertainty. Search snippets are not
evidence. No implementation may start from a digest or an unapproved child
plan.

Phase 2A produces only run-scoped artifacts:

- one validated JSON artifact per successful quest output;
- one deterministic digest JSON file;
- one parent issue Markdown preview;
- one child issue Markdown preview only for each material-change result.

No Phase 2A workflow creates or updates an issue, comment, branch, pull request,
workflow dispatch, schedule, or content file.
2 changes: 2 additions & 0 deletions .github/agentic-refresh/tooling.json
Loading