| name | frame-credit-emission |
| description | Frame credit row emission and deferred credit-input methodology for agents and CE Gate surfaces. Use when an agent or skill needs to write frame credits to the PR-body pipeline-metrics block, or when a pre-PR agent needs to persist a credit-input marker for later harvest. DO NOT USE FOR: predicate evaluation logic (use frame port adapter files), or builder function implementation (use frame-credit-ledger-core.ps1). |
Frame Credit Emission
Reusable methodology for frame credit row emission across all 17 frame ports. Defines the terminal-step contract, credit-input marker schema, and locus-category routing for agents and skills that contribute to the frame ledger.
When to Use
- When an agent completes a terminal step and must write a credit row to the PR-body pipeline-metrics block
- When a pre-PR agent (Experience-Owner, Solution-Designer, Issue-Planner) must persist a credit-input marker for deferred emission
- When Code-Conductor harvests credit-input markers at PR-creation time
- When CE Gate orchestration evaluates per-surface applicability and emits credits for each surface independently
Locus Categories
Frame credit emission follows one of four locus categories based on the port's lifecycle position relative to PR creation:
agent-post-pr
Ports owned by specialist agents whose terminal step occurs after PR creation. The agent writes its credit row directly into the <!-- pipeline-metrics --> block on the PR body via the appropriate Build-*CreditRow function.
Ports in this category: implement-code, implement-test, implement-refactor, implement-docs, process-review
Terminal-step contract: Read the port's adapter file from frame/ports/{port}.yaml, evaluate the applies-when predicate against the changeset, call the matching Build-*CreditRow function with port-specific evidence, and upsert the credit row into the PR-body pipeline-metrics block.
Dedupe contract (process-review only): Build-ProcessReviewCreditRow is the canonical emitter. If a process-review credit row already exists in the pipeline-metrics block with status: passed or status: not-applicable, Code-Conductor skips emission (additive-merge rule, D9). Note: the SMC-16 not-persisted synthesis path covers only the review port (judge-sentinel-driven); it does not extend to process-review — no process-review-specific sentinel exists.
Senior Engineer skill-as-adapter credits
For adapter-type: work skill adapters, the default executor is agents/Senior-Engineer.agent.md. Spine-Runner dispatches Senior Engineer with the planner-selected adapter path, and the Senior Engineer subagent emits the terminal credit row after validation; Spine-Runner verifies the row but does not write it directly.
Credit attribution remains builder-owned. Existing Build-*CreditRow functions are unchanged; the work adapter passes its repo-relative adapter path as the builder adapter value and repeats that path in human-readable evidence. For Build-ImplementCodeCreditRow, the forward-compatible row key set guarded for #557 is exactly port, adapter, status, evidence. The #557 schema migration should preserve this set-equal assertion until the canonical schema intentionally changes.
agent-pre-pr
Ports owned by pipeline-entry agents that complete before any PR exists. These agents use a two-stage deferred-emission pattern:
- Stage A (agent terminal step): Post a credit-input comment alongside the existing completion marker, with shape
<!-- credit-input-{port}-{ID} --> <yaml>...</yaml>. The YAML payload carries the port-specific evidence the builder needs.
- Stage B (PR-creation harvest): Code-Conductor reads the linked issue's comments at
gh pr create time, locates <!-- credit-input-{port}-{ID} --> blocks for each pipeline-entry port, and calls the matching Build-*CreditRow to construct the actual credit rows for the PR-body pipeline-metrics block.
Ports in this category: experience, design, plan
Terminal-step contract: Post a credit-input marker comment with the required YAML payload shape (see Credit-Input Marker Schema below). Code-Conductor harvests at PR-creation time.
skill-only
Ports owned by skills rather than agents. Code-Conductor owns the selector and invokes the appropriate builder function with evidence from the skill's execution.
Ports in this category: post-pr, review
Terminal-step contract: Code-Conductor reads the port's adapter file from frame/ports/{port}.yaml and emits the skill-owned credit row for that port. For post-pr, call Build-PostPrCreditRow with the post-merge checklist outcomes from the post-pr-review skill. For review, follow the review-credit-emission reference, call Build-ReviewCreditRow, and use review-specific evidence such as judge ruling status, reviewed PR context, and persisted review ledger or sentinel details. Upsert the resulting credit row into the PR-body pipeline-metrics block.
pr-body-pipeline-metrics
Ports emitted into the PR-body <!-- pipeline-metrics --> block by Code-Conductor or the owning skill after PR creation, outside the specialist-agent terminal-step path.
Ports in this category: release-hygiene, post-fix-review
Terminal-step contract: Verify the port row in credits[]. release-hygiene uses the symmetric-bump state-file contract from the plugin-release-hygiene skill. post-fix-review uses the review-triggered adversarial-review post-fix adapter and may be explicitly skipped when the review trigger is absent.
deferred-skill-only
Formalized trigger-conditional ports whose producer is intentionally deferred. Code-Conductor emits a not-applicable deferred row with Build-DeferredPortCreditRow.
Ports in this category: process-retrospective
Terminal-step contract: Verify a credits[] row with status: not-applicable and evidence beginning with DEFERRED(#NNN):; the port remains excluded from the coverage denominator until its trigger-status becomes live.
ce-gate-per-surface
CE Gate surface ports, where each surface (CLI, browser, canvas, API) is evaluated independently. Code-Conductor delegates the CE Gate phase to Experience-Owner; for each surface, the customer-experience skill's CE Gate orchestration evaluates the surface-touch predicate and either exercises the surface or emits not-applicable.
Ports in this category: ce-gate-cli, ce-gate-browser, ce-gate-canvas, ce-gate-api
Terminal-step contract: Each Build-CeGateCreditRow -Surface {name} -Step {terminal-step-id} call is its own terminal step with its own integrity check. See "Per-Surface Terminal-Step Contract" below for orchestration-failure handling.
Port -> Locus Inference
Spine-Runner consumes this table as the authoritative port-to-locus mapping. Add rows in this order so inferred frame walking remains deterministic.
auto-na and explicit-skip are adapter variant suffixes (unified convention: {port}-auto-na-adapter.md / {port}-explicit-skip-adapter.md), not canonical port rows. Do not add them to this table.
frame-credit-ledger and pre-pr-format are descriptive locus-category aliases from design discussion. They refer to skill-only methodology that runs alongside the ports above, are not canonical ports in frame/ports/*.yaml, and Spine-Runner does not directly verify them.
Credit-Input Marker Schema
Pre-PR agents (agent-pre-pr locus category) use credit-input markers for deferred emission. The marker shape is:
<!-- credit-input-{port}-{ID} -->
port: { port }
adapter: { adapter }
evidence: "{human-readable evidence string}"
Field requirements:
port: The frame port identifier (e.g., experience, design, plan).adapter: The adapter reference that produced this credit. Spine-Runner v2 frames use the repo-relative adapter path; legacy pipeline-entry emitters may continue to use builder adapter names such as work-adapter or auto-na until migrated.evidence: A flat quoted string describing what the agent observed (e.g., "issue #123; plan-issue marker posted"). The harvester passes this string verbatim as the -Evidence parameter to the matching Build-*CreditRow function. Do not use a nested YAML mapping here - the harvester's flat key-value parser will silently drop nested fields.
Persistence rule: Persist the credit-input marker as a GitHub issue comment immediately after the agent's completion marker comment, via skills/session-memory-contract/scripts/persist-marker.ps1 (family credit-input, post-new, with a YAML-shape validator adapter enforcing the port/adapter/evidence field requirements above pre-write) — the ONLY documented write path for this marker; never a hand-composed gh issue comment call. What that rule buys is a single audited writer — not protection from updated_at advancement. This family is post-new, so the write POSTs a fresh comment and advances nothing; the hazard arrives when a write replaces a whole body already carrying someone else's marker. See skills/session-memory-contract/references/handoff-markers.md § What the write-path rule buys for the write-shape split. See § Burst Persistence in the emitting agent's own body (Experience-Owner, Solution-Designer, Issue-Planner, Code-Conductor) for the canonical-order burst-manifest invocation that includes this write. Code-Conductor harvests all credit-input markers at PR-creation time by reading the issue's comments and calling the matching builder functions.
Additive-merge rule: If the PR-body pipeline-metrics block already contains a credit row for a given port (e.g., from a prior credit-write pass), Code-Conductor skips harvesting the credit-input marker for that port. No double-write, no overwrite.
Per-Surface Terminal-Step Contract
CE Gate surfaces are evaluated independently. Each surface's credit row is its own terminal step:
- Predicate evaluation: For each surface (
cli, browser, canvas, api), evaluate the surface-touch predicate (changeset.touches{Surface}Surface()).
- Surface exercise or N/A: If the predicate is true, exercise the surface and capture evidence. If false, emit
status: not-applicable.
- Credit emission: Call
Build-CeGateCreditRow -Surface {name} -Step {terminal-step-id} with the surface-specific evidence and upsert the credit row.
- Orchestration-failure handling (planned - wrapper not yet implemented): when the orchestration wrapper ships, a crash after partial evaluation will cause it to emit the remaining-surface credits as
status: inconclusive with block_kind: orchestration and evidence: "orchestration crashed before surface evaluated". Until then, manually emit missing-surface rows on crash.
Cycle-aware emission uses (port, terminal-step-id) as the additive-merge identity for positive terminal steps. Every Build-*CreditRow function accepts optional -Step; omit it or pass 0 only for legacy plans and spine-omitted: plan-too-small plans, where the row keeps the legacy (port, 0) identity. For a spine-backed plan, pass the positive step number from the terminal slice marked for that port. Multiple rows for the same port may coexist only when their positive terminal-step-id values differ; an existing row with the same (port, terminal-step-id) wins and must not be overwritten.
Deferred-Port Emission Convention
Ports whose trigger predicate is formalized but deferred to a future issue use Build-DeferredPortCreditRow instead of the port-specific builder. The evidence string always begins with DEFERRED(#NNN): — this prefix is the migration-detection contract (regex ^DEFERRED\(#\d+\):).
When to use: the port's YAML declares trigger-status: deferred and applies-when: never. The port is excluded from the coverage denominator until trigger-status flips to live in the producing issue.
Example (process-retrospective, deferred to #348):
port: process-retrospective
adapter: explicit-skip
status: not-applicable
evidence: "DEFERRED(#348): trigger predicate deferred to #348 (since 2026-05-03); port excluded from coverage denominator until trigger-status flips to live."
Pre- vs post-#348 distinction: rows with the DEFERRED(#NNN): prefix are visually distinguishable from post-cutover rows (which will carry real trigger evidence). Migration scripts detect the prefix and convert deferred rows to live rows when #348 ships its producer.
Related
Gotchas
- Pre-PR agents (Experience-Owner, Solution-Designer, Issue-Planner) cannot write PR-body credits directly because no PR exists yet. They must post a
<!-- credit-input-{port}-{ID} --> marker with the required YAML payload. Code-Conductor harvests these markers at PR-creation time and calls the matching builder functions to populate the pipeline-metrics block.
Frame Ports Filled By This Skill
None. This is supporting methodology only. This skill declares no provides: field and fills no frame port. It documents the emission contract for agents and skills that do fill frame ports.
