Field reports

Skill Field Reports

SDMCC-safe field-report categories, allowed data, forbidden data, and operational sorting guidance.

Skill Field Reports

Skill field reports are governed operational evidence about how a skill behaved in real use. They are adapted from the ClawdMarks "field report" idea, but SDMCC treats them as signed-in workspace signal, not public social reviews.

Goals

Help the next teammate or agent understand whether a skill is ready to use.
Surface missing context, permission problems, validation failures, install friction, and unsafe ambiguity.
Give reviewers evidence for promotion, docs fixes, and backlog follow-up.
Keep reports small, sanitized, and tied to SDMCC-owned skill identifiers.

Categories

Category	Use when	Operational meaning
`worked_well`	The skill completed the expected workflow with useful output.	Positive evidence for adoption or promotion.
`missing_context`	Required project, board, parent item, environment, or policy context was absent.	The skill or reference docs may need clearer prerequisites.
`permission_blocked`	Azure DevOps, repo, AWS, Cognito, SharePoint, or another approved system blocked a step.	Access or role setup needs review.
`validation_failed`	A repo check, fixture, parser, smoke test, or review scenario failed.	The skill or platform code needs correction before promotion.
`install_friction`	Install commands, host setup, CLI extension, certificate trust, or target path setup caused problems.	Install docs or setup scripts need improvement.
`unsafe_ambiguity`	The agent could not proceed safely without risking the wrong board, repo, secret, environment, or external action.	Instructions need a stronger confirmation or routing rule.
`suggested_improvement`	The run worked, but the reporter found a reusable improvement.	Candidate backlog or documentation update.

Allowed Fields

Field	Classification	Rule
`skillId`	Allowed	Store the repo skill folder name or stable skill identifier.
`category`	Allowed	Store one of the approved categories above.
`rating`	Allowed	Store whole numbers from 1 to 5 as a coarse usefulness signal.
`comment`	Controlled	Store a sanitized short comment, capped by the implementation. No raw logs or source text.
`submittedAt`, `createdAt`, `updated`	Allowed	Store ISO timestamps for sorting and retention.
`submittedBy`	Controlled	Use only for authenticated workspace display and authorization context. Do not copy into public pages.
`source`	Allowed	Store `hosted`, `local`, `fixture`, or another approved source label.

Forbidden Data

Do not store or display these in field reports:

PATs, passwords, bearer tokens, API keys, client secrets, auth headers, Cognito tokens, AWS credentials, or dashboard API keys.
.env*, .mcp.json, local auth files, raw deployment logs, generated local app state, screenshots with secrets, or raw exports.
Raw prompts, raw transcripts, meeting recordings, full Azure Boards comments, work item descriptions, acceptance criteria excerpts, or raw invalid JSON payloads.
Local filesystem paths, machine names, OS user profile details, or private browser/session state.
iDPCC, ClawdMarks, or other source-project identifiers unless a repo doc clearly labels them as source-history context.

Sorting And Filters

Current SDMCC data supports these library sorts:

Needs Review: puts submitted, in-review, experimental, deprecated, or issue-heavy skills first.
Recently Updated: uses the skill updated date from generated inventory.
Production Ready: uses the repo status lifecycle from the root index and skill README.
Most Reported Issues: uses field-report issue categories and low ratings.

Planned but not yet supported:

Most Installed/Used: requires a governed install-completion or usage event store. Do not infer installs from copy-button clicks, local browser state, or repo presence.

Implementation Notes

The existing skill-feedback storage path is the governed SDMCC field-report path. The implementation should:

Preserve the existing rating and comment behavior for older reports.
Add category as an optional, validated field.
Summarize category counts and issue counts for sorting without returning raw comments to public surfaces.
Keep hosted reports behind Cognito-authenticated APIs.
Keep local reports in browser storage for development only.
Use fixtures and tests for category validation, issue-count summaries, and operational sort behavior.

Example Fixture

{
  "skillId": "azure-devops-pbi",
  "category": "validation_failed",
  "rating": 2,
  "comment": "Dry-run validation found a missing parent Feature confirmation step.",
  "source": "fixture"
}