CtrlK
BlogDocsLog inGet started
Tessl Logo
PluginPublic

adobe/aem-edge-delivery-services

Skills for building AEM Edge Delivery Services sites — site creation, block development, content modeling, code review, testing, and page import.

Quality

91%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Low

Low-risk findings worth noting

Overview
Quality
Evals
Security
Files

6-reflect.mdskills/snowflake/phases/

Phase 6 — Reflect

Goal: capture findings from this run so the next run is better. This phase does NOT close the iteration — closure is a separate, user-initiated action.

6.1 — Update the project's notes.md

Append a ## Phase: Round-trip and (if relevant) ## Phase: Reflect section summarizing what happened in each phase. Include:

  • Local round-trip outcome (sections present, slot count, console errors, screenshot inventory)
  • Production round-trip outcome (URLs, sanity-probe codes, Media Bus behavior)
  • Anything that surfaced as a project-specific finding (e.g. "this source had a unique X pattern")
  • Anything that surfaced as a CROSS-PROJECT finding (e.g. "a new pipeline behavior we should encode in the methodology")

6.2 — Update the project's learnings.md

Create or append to <projectsDir>/<NNN>-<slug>/learnings.md. Each entry is dated ## YYYY-MM-DD — short title with sections:

  • Context (one paragraph)
  • Visible symptom (if a bug)
  • Fix applied to this project
  • Generic rule (if promotable)

Mark cross-project items with [promoted] so a reader can scan for the ones that landed in the skill knowledge.

6.3 — Promote cross-project findings

For every finding marked [promoted]:

  1. Add an entry to the target repo's experiments/knowledge/learnings.md (the project-local cross- project learnings file). Top of the file, dated, with full context.
  2. If the finding is a rule the methodology should enforce, update the target repo's experiments/knowledge/methodology.md — specifically the phase the rule applies to (Generate / Round-trip / etc.).
  3. If the finding contradicts or extends the substrate documentation, update experiments/knowledge/architecture.md too.

These updates land in the SAME commit that closes the run, when the user asks for closure. Until then, they're staged but uncommitted.

Promote to the skill bundle itself

If the finding is generic enough to live in the skill (not project- specific), it should also be PR'd to the skill repo (this repo). The agent CAN propose this via a separate worktree on the skill repo — or, if that's friction, just collect candidates in notes.md under a "promote-to-skill" heading and let the user raise the PR when convenient. Hosts that don't make git cheap can default to the latter.

6.4 — Run a timing summary (optional but helpful)

If the run captured per-phase timestamps in state.json, summarize:

phase             duration    notes
capture           <s>
analyze           <s>
generate          <s>         largest single phase, usually
wire              <s>
roundtrip-local   <s>
roundtrip-prod    <s>
reflect           <s>

Add this to notes.md under a "Timings" section. Helps tune future runs.

6.5 — Auto-memory updates (host-dependent)

If the host supports persistent memory across sessions (Claude Code auto-memory, similar primitives elsewhere) and the run surfaced a user-feedback rule (something about the user's preferences, not about the conversion), save it. Examples from prior runs:

  • "User wants closure to be explicit, not agent-initiated."
  • "User wants to be asked before phases are skipped."

Skip this on hosts that don't have a memory primitive.

6.6 — Stop. Do NOT close the iteration.

The skill's job ends here. Closure (iter-<NNN>-close tag, fast-forward of the integration trunk) requires the user to say "close this iteration" or equivalent. The agent presents the run as ready-for-close, lists what would happen on close, and waits.

When the user gives the explicit go-ahead, run:

# Resolve branch + trunk + tag conventions from config (with defaults)
BRANCH_PREFIX=$(jq -r '.branchPrefix // "snowflake-"' \
  .snowflake/config.json 2>/dev/null || echo "snowflake-")
TRUNK=$(jq -r '.trunkBranch // "main"' \
  .snowflake/config.json 2>/dev/null || echo "main")
TAG_PREFIX=$(jq -r '.tagPrefix // "snowflake-"' \
  .snowflake/config.json 2>/dev/null || echo "snowflake-")

BRANCH="${BRANCH_PREFIX}${NNN}"
TAG="${TAG_PREFIX}${NNN}-close"

git checkout "$BRANCH"
git tag "$TAG"
git checkout "$TRUNK"
git merge --ff-only "$BRANCH"
git push origin "$TRUNK"
git push origin "$BRANCH"
git push origin "$TAG"

# Update the project README's Status section to:
#   "Closed YYYY-MM-DD (tag $TAG)"

The defaults match a typical EDS site repo (main as trunk, snowflake/NNN as run branches, snowflake-NNN-close as closure tags). The R&D project the skill was distilled from used different conventions; if a repo wants those, set them in .snowflake/config.json.

Then update state.json one last time: state.phase = "closed", state.phaseStatus = "complete", state.closedAt = "<timestamp>".

skills

README.md

tile.json