Migrated from Method backlog

This issue was created from a legacy filesystem backlog card. GitHub Issues are now the live work tracker; repository docs remain Method evidence.

Source backlog: docs/method/backlog/cool-ideas/METHOD_leash-files.md
Original lane: cool-ideas
Original legend: METHOD

Original backlog card

Leash files: a structural lane for temporary scaffolds

Legend: METHOD

The problem

Cross-repo or cross-cycle work routinely produces temporary scaffolds:
client-side mirrors of unfinished engine APIs, transitional adapters,
"good for now" boundary shims. They are deliberate decisions, not
mistakes. The failure mode is not introducing them — it is forgetting
to delete them when their reason expires.

The natural pressure on a working scaffold is to accrete features,
become depended on, and resist removal. Prose-only backlog cards
("we'll remember to delete this") rely on humans being the system
that nags. Humans are goldfish with keyboards.

The 2026-05-30 jedit PR flyingrobots/echo#33 review-resolution session ended with
exactly this concern about JeditWorldlineSessionPort and friends
("Temporary things love to buy furniture. Don't let it."). The
prototype response — jedit/docs/method/backlog/asap/sessions-migration.md
— is prose-heavy and human-friendly but not machine-naggable.

The shape

Promote a new backlog lane:

docs/method/backlog/leash/<scaffold-slug>.md

Each leash file is a small structured record naming a scaffold, the
event that should free it, and the symbols whose disappearance is the
deletion proof. The lane sits alongside asap/, bad-code/,
cool-ideas/, and graveyard/; it does not replace any of them. A
leash and a prose asap/ card cross-reference each other when both
exist.

Required frontmatter

---
scaffold: JeditWorldlineSessionPort
repo: jedit
introduced_by:
    pr: 33
    merged_sha: 90245c73
    date: 2026-05-30
reason: |
    Engine-side Session/WorldlineId types do not exist yet; the client
    carries a transitional mirror so the EINT cutover can ship without
    blocking on echo cycle 0025 Phase 2.
deletion_trigger:
    repo: echo
    cycle: "0025-sessions-as-causal-contexts"
    phase: "Phase 2 GREEN"
symbols:
    - JeditWorldlineSessionPort
    - JeditTransportSeam
    - createInMemoryJeditWorldlineSessionPort
    - installed-jedit-eint-bridge
status: active
companion_cards:
    - jedit/docs/method/backlog/asap/sessions-migration.md
---

Required fields: scaffold, repo, introduced_by, reason,
deletion_trigger, symbols, status. Optional but encouraged:
companion_cards, notes body section after the frontmatter for
context.

status is one of active, triggered, deleted, escalated.

Lifecycle

1. active — the scaffold is in production, the trigger has not
   fired. The leash exists. No action.
2. triggered — the deletion trigger condition has been met
   (e.g. the named cycle/phase closed). Tooling moves the file into
   triggered and surfaces it on the next cycle-close summary.
3. deleted — the named symbols are gone from the repo. Tooling
   confirms via grep and moves the leash file into
   docs/method/graveyard/ with the deletion SHA appended.
4. escalated — the deletion trigger fired but the symbols
   still exist after a defined grace window. Tooling promotes the
   leash to asap/ and tags the next cycle.

What tooling has to do

This card lands the lane registration so existing tooling sees
leash files from day one:

• crates/method/src/workspace.rs LANES includes "leash" so
  cargo xtask method status counts leash files alongside other
  backlog lanes.
• crates/method/src/graph.rs GRAPH_LANES includes "leash"
  and lane_colors() gives it an amber + dark-orange palette
  ("tied, on a clock" — distinct from asap's soft yellow).
• crates/method/tests/status_tests.rs includes
  status_counts_leash_lane_files as a regression so a future
  refactor cannot silently drop the lane.

The deeper trigger machinery is proposed as a follow-up:
xtask leash-audit runs in CI / locally and:

• Parses every docs/method/backlog/leash/*.md frontmatter.
• For each active leash:
  • If the deletion_trigger.cycle has closed (look at
    docs/method/retro/<cycle>/), promote to triggered.
• For each triggered leash:
  • Greps the leash's repo for every entry in symbols. If zero
    hits, promote to deleted and move to graveyard. If any hits,
    surface as a warning.
• For triggered leashes older than the configured grace window
  (default: 1 cycle), promote to escalated.

The audit script is dumb on purpose. The hard work is keeping the
leash records honest, not the matching.

Why this is the right size of mechanism

• It does not introduce a new project-management tool. It is a
  backlog lane, same as every other lane.
• It does not require humans to remember anything. The trigger
  condition is encoded; tooling notices.
• It scales: a repo with five active leashes still produces a
  five-line audit summary.
• It composes with asap/: prose explanations stay where prose
  explanations belong; structural records stay where structural
  records belong.
• It composes with graveyard/: completed leashes become a
  historical record of "we said this was temporary, here is the SHA
  that proved it."

Prototype

The first concrete leash, landing alongside this card, is in the
jedit repo at:

jedit/docs/method/backlog/leash/jedit-session-port.md

It names JeditWorldlineSessionPort + JeditTransportSeam +
createInMemoryJeditWorldlineSessionPort + the installed EINT
bridge as scaffolds whose deletion is bound to Echo cycle 0025
Phase 2 GREEN.

Not in scope here

• The xtask leash-audit implementation (the trigger / symbol /
  status transition machinery). This card ships the lane
  registration so leash files are visible to existing tooling from
  day one; the audit gets built when the second leash file shows
  up (the universal "you only need a Tool when you have two of the
  thing" rule).
• Cross-repo automation that opens deletion PRs automatically. That
  is the natural next step but adds blast radius; ship the manual
  audit first.

Companion memory

The 2026-05-30 session captured the leash-pattern motivation as a
project memory in the agent's own store; see
project-jedit-session-port-leash in the agent's memory index. That
memory and this card are the same idea at two altitudes: the memory
keeps the agent honest in casual touches of the codebase; this card
keeps METHOD honest at the cycle/process level.