x-internal-build-story-plan
Orchestrates parallel story-planning (Phase 1 carve-out of x-implement-story): invokes x-plan-architecture (Step 1A) and then dispatches 5 sibling Agent subagents in ONE assistant message for implementation plan, test plan, task breakdown, security assessment, and compliance assessment (Steps 1B-1F). Applies the Rule 13 SUBAGENT-GENERAL pattern as the canonical parallel-planning gateway. Scope-aware: SIMPLE skips 1E/1F. Returns a consolidated envelope of artifact paths to the calling orchestrator. Fifth skill in the x-internal-* convention and the second under internal/plan/ (after x-internal-load-story-context).
- License 许可
- MIT
- 最近更新
- 2026-10-27
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- x-internal-build-story-plan
- category
- internal-plan
- visibility
- internal
- description
- Orchestrates parallel story-planning (Phase 1 carve-out of x-implement-story): invokes x-plan-architecture (Step 1A) and then dispatches 5 sibling Agent subagents in ONE assistant message for implementation plan, test plan, task breakdown, security assessment, and compliance assessment (Steps 1B-1F). Applies the Rule 13 SUBAGENT-GENERAL pattern as the canonical parallel-planning gateway. Scope-aware: SIMPLE skips 1E/1F. Returns a consolidated envelope of artifact paths to the calling orchestrator. Fifth skill in the x-internal-* convention and the second under internal/plan/ (after x-internal-load-story-context).
- allowed-tools
- Bash, Skill, Agent
- argument-hint
- --story-id <story-XXXX-YYYY> --epic-id <XXXX> [--scope <SIMPLE|STANDARD|COMPLEX>] [--skip-review]
- context-budget
- medium
- user-invocable
- false
- requires-capabilities
- []
02 资源文件 1
- 参考文档references/full-protocol.md11.4 KB
03 SKILL.md 正文
Global Output Policy
- Language: English ONLY.
- Tone: Technical, Direct, and Concise.
- Efficiency: Remove all conversational fillers and greetings to save tokens.
🔒 INTERNAL SKILL Esta skill é invocada apenas por outras skills (orquestradores). NÃO é destinada a invocação direta pelo usuário. Caller principal: x-implement-story (Phase 1 carve-out). Quinta skill da convenção
x-internal-*(após x-internal-update-status pilot 0049-0005, x-internal-write-report 0049-0006, x-internal-normalize-args 0049-0007, e x-internal-load-story-context 0049-0011). Segunda skill na subdirinternal/plan/(co-habita com x-internal-load-story-context). A subdirplan/agrupa as skills que orquestram lógica de planejamento; difere deinternal/ops/, cujas sibling skills mutam estado (execution-state.json, reports).
Skill: x-internal-build-story-plan
Purpose
Carve out Phase 1 (Architecture Planning + Parallel Planning) of
x-implement-story into a single orchestration skill. The ~250
inline lines currently inside x-implement-story Phase 1 become a
single Skill(skill: "x-internal-build-story-plan", …) invocation;
the orchestrator shrinks to a read-the-envelope consumer.
Responsibilities (single):
- Invoke
x-plan-architecturefor the story (Step 1A). - Dispatch 5 sibling
Agent(general-purpose, …)subagents in ONE assistant message for Steps 1B (implementation plan), 1C (test plan), 1D (task breakdown), 1E (security assessment), and 1F (compliance assessment). - Gate Steps 1E and 1F behind
--scope(SIMPLE → skip 1E/1F). - Collect artifact paths from every subagent, assemble a consolidated response envelope, and emit it on stdout.
- Translate individual subagent or
x-plan-architecturefailures into the stable exit-code catalogue below.
Non-responsibilities (explicit):
- The skill does NOT read, re-parse, or classify the story file —
that is
x-internal-load-story-context's contract (story-0049-0011). - The skill does NOT mutate
execution-state.json, story**Status:**headers, or IMPLEMENTATION-MAP — those belong tox-internal-update-status(story-0049-0005). - The skill does NOT run the parallelism-collision gate — Phase 1.5
remains the orchestrator's responsibility via
x-evaluate-parallelism(EPIC-0041). - The skill does NOT perform pre-check / freshness classification of
the 7 planning artifacts — that runs upstream inside
x-internal-load-story-context(Step 4). Callers only invoke this skill whenplanningMode ∈ {HYBRID, INLINE}and a regen is actually required.
Convention Anchors (x-internal-*)
| Aspect | Value | Rationale |
|---|---|---|
| Path | internal/plan/x-internal-build-story-plan/ |
internal/ prefix scopes visibility; plan/ co-locates with the sibling read-only carve-out (x-internal-load-story-context) |
Frontmatter visibility |
internal |
Generator filters these from /help menu |
Frontmatter user-invocable |
false |
Declarative complement to visibility: internal |
| Body marker | > 🔒 **INTERNAL SKILL** block as first non-frontmatter content |
Visible to humans browsing the repo; no parsing required |
| Allowed tools | Bash, Skill, Agent |
Minimal for an orchestrator: Skill for Step 1A (x-plan-architecture), Agent for Steps 1B-1F, Bash for argv parsing and the envelope jq -nc assembly |
| Naming | x-internal-{subject}-{action} |
Mirrors Rule 04 skill taxonomy; story-build-plan = subject+action |
Audit rule: Rule 22 (Lifecycle Integrity) validates every skill under
internal/** satisfies all 6 anchors above. Violations fail
LifecycleIntegrityAuditTest.
Triggers
Bare-slash form is intentionally omitted — this skill is never
invoked by a human typing /x-internal-build-story-plan in chat.
All invocations follow Rule 13 INLINE-SKILL pattern from a calling
orchestrator:
Skill(skill: "x-internal-build-story-plan",
args: "--story-id story-0049-0012 --epic-id 0049 --scope STANDARD")
Parameters
| Parameter | Required | Default | Description |
|---|---|---|---|
--story-id <id> |
M | — | Story identifier (story-XXXX-YYYY canonical form) |
--epic-id <id> |
M | — | 4-digit epic identifier (XXXX) — used to resolve ai/epics/epic-XXXX/ |
--scope <tier> |
O | STANDARD |
SIMPLE / STANDARD / COMPLEX; SIMPLE skips Steps 1E (security) and 1F (compliance) |
--skip-review |
O | false |
When true, signals downstream subagents to omit inline peer-review passes in the generated artifacts (advisory only — does not affect artifact count) |
All four argument forms (--key value, --key=value, empty, and
unknown-flag rejection) follow the Rule-14 rejection matrix: reject
unknown flags and missing required flags with exit 64 (sysexits
EX_USAGE). The --story-id value is normalised to lowercase and
validated against ^story-[0-9]{4}-[0-9]{4}$; --epic-id is
zero-padded to 4 digits.
Response Contract
On success the skill writes a single-line JSON object to stdout:
| Field | Type | Always Present | Description |
|---|---|---|---|
archPlan |
String (path) |
yes | Path of the architecture plan (ai/epics/epic-XXXX/plans/arch-story-XXXX-YYYY.md) |
implPlan |
String (path) |
yes | Path of the implementation plan (plan-story-XXXX-YYYY.md) |
testPlan |
String (path) |
yes | Path of the test plan (tests-story-XXXX-YYYY.md) |
taskBreakdown |
String (path) |
yes | Path of the task breakdown (tasks-story-XXXX-YYYY.md) |
securityAssessment |
String (path) | null |
no — null when scope=SIMPLE |
Path of the security assessment (security-story-XXXX-YYYY.md) |
complianceAssessment |
String (path) | null |
no — null when scope=SIMPLE |
Path of the compliance assessment (compliance-story-XXXX-YYYY.md) |
taskMap |
String (path) |
yes | Path of the task-implementation map (task-implementation-map-story-XXXX-YYYY.md) |
scope |
String |
yes | Echo of the resolved scope tier — lets the caller verify the gate applied |
skipped |
Array<String> |
yes | Names of gated steps skipped (e.g., ["1E", "1F"] in SIMPLE; [] otherwise) |
The six artifact basenames (all under ai/epics/epic-XXXX/plans/):
arch-story-XXXX-YYYY.md(Step 1A —x-plan-architecture)plan-story-XXXX-YYYY.md(Step 1B — Senior Architect subagent)tests-story-XXXX-YYYY.md(Step 1C —x-plan-testssubagent)tasks-story-XXXX-YYYY.md(Step 1D —x-lib-decompose-tasksubagent)security-story-XXXX-YYYY.md(Step 1E —x-model-threatssubagent; omitted in SIMPLE)compliance-story-XXXX-YYYY.md(Step 1F — Compliance Engineer subagent; omitted in SIMPLE)
The task-implementation-map (task-implementation-map-story-XXXX-YYYY.md)
is produced by the Step 1D subagent as part of the task-decomposition
artifact (EPIC-0038 v2 contract).
Exit Codes
| Code | Name | Condition | Message Format |
|---|---|---|---|
| 0 | SUCCESS | All required artifacts produced | — |
| 1 | ARCH_PLAN_FAILED | x-plan-architecture (Step 1A) returned non-zero or failed to produce arch-story-*.md |
Architecture plan failed: <detail> |
| 2 | SUBAGENT_FAILED | Any of Steps 1B-1F (subagent dispatch) failed or did not produce its artifact | Subagent <name> failed: <detail> |
| 64 | EX_USAGE | Unknown flag, missing required flag, or malformed --story-id / --epic-id |
usage: <detail> |
No partial-success state: if any required step fails, the skill exits non-zero. Optional steps (1E / 1F in SIMPLE) are not errors when skipped by the gate; they ARE errors when dispatched and failed.
Workflow
Step 1 — Argument parsing and path resolution
Parse arguments with the Rule-14 single-file while (($#)) loop;
reject unknown flags / missing required flags / empty values with
exit 64. Normalise --story-id to lowercase and verify against
^story-[0-9]{4}-[0-9]{4}$; zero-pad --epic-id to 4 digits.
Derive:
epic_dir="ai/epics/epic-${epic_id}"
story_file="${epic_dir}/${story_id}.md"
plans_dir="${epic_dir}/plans"
When ${epic_dir} is not a directory, exit 64 with
usage: epic dir not found: ai/epics/epic-<id>. The skill does NOT
re-validate the story file existence — that is the caller's
responsibility (x-internal-load-story-context runs upstream).
Step 1A — Architecture plan (sequential)
Invoke x-plan-architecture via the Rule 13 INLINE-SKILL pattern:
Skill(skill: "x-plan-architecture",
args: "--story-id ${story_id} --epic-id ${epic_id}")
Expected output: ${plans_dir}/arch-story-${story_id}.md.
On any failure (non-zero exit from x-plan-architecture, or the expected
file does not exist after invocation), emit to stderr:
Architecture plan failed: <underlying message> and exit 1.
Step 1A MUST complete before Step 1B-1F dispatch: later steps read architectural context (ports, layers, dependency direction) from the arch plan.
Steps 1B-1F — Parallel subagent dispatch (one assistant message)
Dispatch exactly 5 sibling Agent(general-purpose, …) tool calls
in ONE assistant message — the canonical Rule 13 Pattern 2
(SUBAGENT-GENERAL) parallel-launch shape. When --scope=SIMPLE,
dispatch only Steps 1B / 1C / 1D (3 siblings) and record 1E and
1F in skipped.
Each subagent prompt follows the structure:
FIRST ACTION: <no-op — TaskCreate/TaskUpdate unavailable in this harness>.
You are a <role>. Read context files:
- ai/epics/epic-${epic_id}/${story_id}.md (story)
- ai/epics/epic-${epic_id}/plans/arch-story-${story_id}.md (arch plan)
- <role-specific templates under .claude/templates/>.
Produce <artifact> at ${plans_dir}/<basename>.
LAST ACTION: return the absolute path of the artifact you produced.
Per-step role and artifact mapping:
| Step | Role | Skill delegate (if any) | Artifact basename |
|---|---|---|---|
| 1B | Senior Architect | — (inline prompt) | plan-story-${story_id}.md |
| 1C | QA Engineer | x-plan-tests via Skill(…) inside subagent |
tests-story-${story_id}.md |
| 1D | Task Decomposer | x-lib-decompose-task via Skill(…) inside subagent |
tasks-story-${story_id}.md + task-implementation-map-story-${story_id}.md |
| 1E | Security Engineer | x-model-threats via Skill(…) inside subagent |
security-story-${story_id}.md |
| 1F | Compliance Engineer | — (inline prompt) | compliance-story-${story_id}.md |
Origin Marker Emission (EPIC-0059 — mandatory for all 1B-1F artifacts)
Every subagent producing a planning artifact MUST prepend the YAML frontmatter block before writing the file content:
---
generated-by: x-internal-build-story-plan@$(git rev-parse HEAD 2>/dev/null || echo "unknown")
generated-at: $(date -u +%Y-%m-%dT%H:%M:%SZ)
story-id: ${story_id}
---
Include the following instruction in each subagent prompt:
MANDATORY — FIRST STEP of writing the artifact: Prepend the following YAML frontmatter block at the very top of the output file before any Markdown content:
--- generated-by: x-internal-build-story-plan@<run: git rev-parse HEAD> generated-at: <run: date -u +%Y-%m-%dT%H:%M:%SZ> story-id: ${story_id} ---This is required by
audit-execution-integrity.shPhase-1 validation (EPIC-0059, Rule 24). Artifacts without this block fail the CI audit withEIE_EVIDENCE_MISSING.
The --skip-review flag propagates into the subagent prompt when
set: a single sentence instructing the subagent to omit the inline
peer-review pass customarily appended to its artifact.
Step 2 — Collect results and assemble envelope
After all subagents return, validate that every expected artifact exists on disk:
for basename in arch plan tests tasks; do
path="${plans_dir}/${basename}-story-${story_id}.md"
[[ -f "${path}" ]] || { err "Subagent ${basename} did not produce ${path}"; exit 2; }
done
Repeat for security- / compliance- ONLY when scope != SIMPLE.
Assemble the response envelope via jq -nc so the shape is
authoritative:
jq -nc \
--arg archPlan "${plans_dir}/arch-story-${story_id}.md" \
--arg implPlan "${plans_dir}/plan-story-${story_id}.md" \
--arg testPlan "${plans_dir}/tests-story-${story_id}.md" \
--arg taskBreakdown "${plans_dir}/tasks-story-${story_id}.md" \
--arg taskMap "${plans_dir}/task-implementation-map-story-${story_id}.md" \
--arg scope "${scope}" \
--argjson skipped "${skipped_json}" \
--arg securityAssessment "${security_path_or_null}" \
--arg complianceAssessment "${compliance_path_or_null}" \
'{archPlan:$archPlan, implPlan:$implPlan, testPlan:$testPlan,
taskBreakdown:$taskBreakdown, taskMap:$taskMap,
securityAssessment: (if $securityAssessment == "" then null
else $securityAssessment end),
complianceAssessment: (if $complianceAssessment == "" then null
else $complianceAssessment end),
scope:$scope, skipped:$skipped}'
Emit on stdout as a single line terminated by \n. Exit 0.
Examples
Example 1 — Happy path: STANDARD scope (6 artifacts)
Skill(skill: "x-internal-build-story-plan",
args: "--story-id story-0049-0012 --epic-id 0049 --scope STANDARD")
Output:
{"archPlan":"ai/epics/epic-XXXX/plans/arch-story-XXXX-YYYY.md","implPlan":"ai/epics/epic-XXXX/plans/plan-story-XXXX-YYYY.md","testPlan":"ai/epics/epic-XXXX/plans/tests-story-XXXX-YYYY.md","taskBreakdown":"ai/epics/epic-XXXX/plans/tasks-story-XXXX-YYYY.md","taskMap":"ai/epics/epic-XXXX/plans/task-implementation-map-story-XXXX-YYYY.md","securityAssessment":"ai/epics/epic-XXXX/plans/security-story-XXXX-YYYY.md","complianceAssessment":"ai/epics/epic-XXXX/plans/compliance-story-XXXX-YYYY.md","scope":"STANDARD","skipped":[]}
Exit: 0.
Example 2 — SIMPLE scope skips 1E / 1F
Skill(skill: "x-internal-build-story-plan",
args: "--story-id story-0049-0001 --epic-id 0049 --scope SIMPLE")
Output (truncated):
{"archPlan":"…","implPlan":"…","testPlan":"…","taskBreakdown":"…","taskMap":"…","securityAssessment":null,"complianceAssessment":null,"scope":"SIMPLE","skipped":["1E","1F"]}
Exit: 0.
Example 3 — COMPLEX scope (same artifact set as STANDARD)
COMPLEX currently produces the same 6 artifacts as STANDARD; the
tier is echoed in scope for downstream decisions (e.g., post-Phase-2
stakeholder review gate in x-implement-story) but does not change
the planning surface.
Example 4 — Architecture plan fails
Skill(skill: "x-internal-build-story-plan",
args: "--story-id story-0049-0099 --epic-id 0049 --scope STANDARD")
Stderr:
Architecture plan failed: x-plan-architecture exit 3 (story not found)
Exit: 1.
Example 5 — Subagent failure
Step 1B (Senior Architect) fails to produce plan-story-…md.
Stderr:
Subagent 1B failed: plan-story-0049-0012.md not produced
Exit: 2.
Outputs
| Artifact | Path | Description |
|---|---|---|
| Response envelope | stdout | Single-line JSON matching the Response Contract |
| Error diagnostic | stderr | Single line, non-empty only on exit ≠ 0 |
| Planning artifacts | ai/epics/epic-${epic_id}/plans/*.md |
Written by x-plan-architecture + Step 1B-1F subagents |
This skill DOES create files — unlike x-internal-load-story-context,
which is strictly read-only. The allowed-tools frontmatter
includes Skill and Agent to reflect this: the writes happen
inside the dispatched subagents / the invoked x-plan-architecture, not
directly from this skill's body.
Error Handling
| Scenario | Action |
|---|---|
| Missing required flag | Print usage: banner to stderr; exit 64 |
| Unknown flag | Print usage: unknown flag … to stderr; exit 64 |
Malformed --story-id |
usage: --story-id must match story-NNNN-NNNN; exit 64 |
| Missing epic dir | usage: epic dir not found: ai/epics/epic-<id>; exit 64 |
x-plan-architecture non-zero |
Architecture plan failed: <detail>; exit 1 |
Expected arch-story-*.md absent after Step 1A |
Architecture plan failed: arch-story-*.md not produced; exit 1 |
| Any subagent returns error | Subagent <step> failed: <detail>; exit 2 |
| Any subagent returns success but required artifact absent | Subagent <step> failed: <basename> not produced; exit 2 |
jq absent on PATH |
Exit 127 with jq is required; abort before dispatch |
--scope value unrecognised |
usage: --scope must be SIMPLE / STANDARD / COMPLEX; exit 64 |
The skill is fail-fast: the first error aborts the run. Partial
artifacts produced by successful subagents remain on disk (the
orchestrator may choose to re-invoke with the same arguments —
x-plan-architecture and each subagent are idempotent with respect to their
artifact path).
Performance Contract
Target: ~3-5 minutes for STANDARD scope on a typical story. The dominant cost is the wall-clock of the slowest subagent (Step 1B-1F run in parallel, so the total ≈ max(individual)). Step 1A is strictly sequential and typically ≤ 60 seconds.
SIMPLE-scope runs save ~0-30 seconds by skipping 1E and 1F (those subagents are typically I/O-bound on template reads, not model inference).
Testing
Acceptance scenarios (mirroring Section 7 of story-0049-0012):
- Happy path — STANDARD. All 6 artifacts produced; envelope
echoes
scope=STANDARD,skipped=[]. - Degenerate — SIMPLE. 4 artifacts produced; envelope reports
securityAssessment=null,complianceAssessment=null,skipped=["1E","1F"]. - Boundary — COMPLEX. Same 6 artifacts as STANDARD; envelope
echoes
scope=COMPLEX. - Error —
x-plan-architecturefails. Exit 1; stderr starts withArchitecture plan failed:. - Error — subagent failure. Exit 2; stderr identifies which
step (
1B/1C/1D/1E/1F).
Coverage requirement: ≥ 95% line / ≥ 90% branch across the
invoking Bash codepaths and the dispatch / envelope-assembly
logic. Goldens (if added) lock the SKILL.md rendering under
src/test/resources/golden/internal/plan/x-internal-build-story-plan/
per the sibling pattern.
Generator Filter Contract
The ia-dev-env generator MUST exclude skills with
visibility: internal from:
- The
.claude/README.mdskill-inventory table. - The
/helpmenu listing surfaced by Claude Code. - User-facing autocomplete in the chat input.
Internal skills are still copied into .claude/skills/ (flat
layout) so Skill(skill: "x-internal-…") invocations from other
skills resolve correctly. The invariant: user cannot see them;
orchestrators can invoke them.
Telemetry
Internal skills DO NOT emit phase.start / phase.end markers —
telemetry is produced by the invoking orchestrator (the phase
wrapping the orchestrator's own step is the correct aggregation
boundary). Passive hooks still capture tool.call for each
underlying Skill / Agent / Bash invocation made from inside
the skill body.
Subagent subagent.start / subagent.end markers around Steps
1B-1F are emitted by the caller (x-implement-story Phase 1)
because that orchestrator owns the enclosing phase. Re-emitting
from inside this skill would double-count the wave.
Reference: Rule 13 (Skill Invocation Protocol), Rule 22 (Lifecycle Integrity Audit), ADR-0010 (Interactive Gates Convention — exempts internal skills from the 3-option menu contract).
Integration Notes
| Skill | Relationship | Context |
|---|---|---|
x-implement-story |
caller (primary) | Phase 1 carve-out: ~250 inline lines collapse to one Skill(skill: "x-internal-build-story-plan", …) invocation in story-0049-0019 |
x-internal-load-story-context |
upstream peer | Caller runs this sibling first to decide whether Phase 1 needs regen (planningMode != PRE_PLANNED); only then invokes x-internal-build-story-plan |
x-plan-architecture |
delegate (Step 1A) | Sequential precondition; produces arch-story-*.md consumed by Steps 1B-1F |
x-plan-tests |
delegate (Step 1C) | Invoked from inside the QA-Engineer subagent |
x-lib-decompose-task |
delegate (Step 1D) | Invoked from inside the Task-Decomposer subagent; produces both the tasks file and the task-implementation-map |
x-model-threats |
delegate (Step 1E) | Invoked from inside the Security-Engineer subagent; skipped in SIMPLE |
x-evaluate-parallelism |
consumer (downstream) | Phase 1.5 of the caller reads the task-implementation-map produced here to compute the collision matrix (EPIC-0041) |
x-internal-update-status |
peer | Separate concern (mutates state); never invoked from this skill |
Downstream stories that depend on this carve-out: story-0049-0019 (orchestrator consumes the envelope and deletes the inline Phase 1 block).
Full workflow detail (subagent prompt catalogue per step, scope-gate
decision table, envelope-assembly edge cases, and the interaction
with x-internal-load-story-context's upstream freshness result)
lives in references/full-protocol.md
per ADR-0011.
04 同仓库更多 Skills
x-parallel-eval
Detects and classifies file-collision risks between work units (tasks, stories, or epic phases) before parallel execution. Reads ## File Footprint blocks produced by x-task-plan / x-story-plan, applies the parallelism-heuristics knowledge pack (hard / regen / soft categories + hotspot overrides), and emits a collision matrix + serialization recommendation in Markdown (default) or JSON.
parallelism-heuristics
Canonical catalog of file-collision heuristics for parallel task, story, and epic execution. Defines the File Footprint block format, the three Conflict Categories (hard / regen / soft), a hotspot list of high-contention paths, and the degrade-with-warning policy. Consumed by x-task-plan, x-story-plan, x-epic-map, x-parallel-eval, and x-dev-*-implement.
x-watch-pr-ci
Polls a PR's CI checks and Copilot review status, blocking until checks complete or timeout. Returns one of 8 stable exit codes (SUCCESS=0, CI_PENDING_PROCEED=10, CI_FAILED=20, TIMEOUT=30, PR_ALREADY_MERGED=40, NO_CI_CONFIGURED=50, PR_CLOSED=60, PR_NOT_FOUND=70). Writes a versioned state-file for session resume.
x-validate-docs
Documentation freshness gate: validates 6 dimensions (readme, api-specs, grpc-proto, adr, skill-docs, system-architecture) against code changes in a PR. Stack-aware — only validates targets declared/auto-detected from ProjectConfig.documentation.targets. Returns exit non-zero on staleness; produces structured report.
x-update-system-architecture
Incrementally updates docs/architecture/system.md after an epic completes. Appends new entries to the Decision Log (§11) using x-internal-write-report --append (dedup by ## ID: marker) and surgically inserts component/integration changes into sections 1-10 via Edit. Idempotent: re-running with the same epic produces a byte-identical system.md.
x-update-architecture
Incrementally updates the service architecture document with changes from architecture plans. Adds new components, integrations, flows, and ADR references without rewriting existing content. Use after implementation to keep architecture documentation current.