@warnyin/sdlc 0.8.0 → 0.9.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (72) hide show
  1. package/CHANGELOG.md +274 -258
  2. package/LICENSE +21 -21
  3. package/README.md +92 -92
  4. package/bin/cli.mjs +682 -682
  5. package/lib/active.mjs +199 -199
  6. package/lib/caps.mjs +46 -46
  7. package/lib/config.mjs +41 -41
  8. package/lib/delta.mjs +227 -227
  9. package/lib/frontmatter.mjs +59 -59
  10. package/lib/glob.mjs +29 -29
  11. package/lib/lenses.mjs +48 -48
  12. package/lib/manifest.mjs +99 -99
  13. package/lib/settings-merge.mjs +63 -63
  14. package/lib/skills.mjs +148 -148
  15. package/lib/validate.mjs +198 -198
  16. package/package.json +42 -42
  17. package/payload/adapters/agents-md.md +8 -8
  18. package/payload/adapters/claude/agents/sdlc-architect.md +12 -12
  19. package/payload/adapters/claude/agents/sdlc-builder.md +14 -14
  20. package/payload/adapters/claude/agents/sdlc-contractor.md +13 -13
  21. package/payload/adapters/claude/agents/sdlc-evaluator.md +13 -13
  22. package/payload/adapters/claude/agents/sdlc-learner.md +16 -16
  23. package/payload/adapters/claude/agents/sdlc-ops.md +11 -11
  24. package/payload/adapters/claude/agents/sdlc-quality.md +13 -13
  25. package/payload/adapters/claude/agents/sdlc-security.md +12 -12
  26. package/payload/adapters/claude/commands/sdlc/converge.md +5 -5
  27. package/payload/adapters/claude/commands/sdlc/init.md +4 -4
  28. package/payload/adapters/claude/commands/sdlc/next.md +4 -4
  29. package/payload/adapters/claude/commands/sdlc/observe.md +4 -4
  30. package/payload/adapters/claude/commands/sdlc/steer.md +4 -4
  31. package/payload/adapters/claude/skills/contract-writing/SKILL.md +26 -26
  32. package/payload/adapters/claude/skills/delta-spec-format/SKILL.md +36 -36
  33. package/payload/adapters/claude/skills/sdlc-conventions/SKILL.md +30 -30
  34. package/payload/adapters/cline.md +8 -8
  35. package/payload/adapters/copilot.md +8 -8
  36. package/payload/adapters/cursor.mdc +7 -7
  37. package/payload/adapters/gemini.md +8 -8
  38. package/payload/adapters/windsurf.md +4 -4
  39. package/payload/hooks/_shared.mjs +138 -138
  40. package/payload/hooks/guard-writes.mjs +87 -87
  41. package/payload/hooks/inject-context.mjs +57 -57
  42. package/payload/hooks/journal.mjs +66 -66
  43. package/payload/hooks/session-summary.mjs +52 -52
  44. package/payload/hooks/validate-artifact.mjs +84 -84
  45. package/payload/playbook/README.md +32 -32
  46. package/payload/playbook/context.md +26 -26
  47. package/payload/playbook/contract.md +29 -29
  48. package/payload/playbook/converge.md +19 -19
  49. package/payload/playbook/design.md +28 -28
  50. package/payload/playbook/init.md +22 -22
  51. package/payload/playbook/lenses.md +64 -64
  52. package/payload/playbook/new.md +41 -33
  53. package/payload/playbook/next.md +24 -24
  54. package/payload/playbook/observe.md +20 -20
  55. package/payload/playbook/principles.md +28 -28
  56. package/payload/playbook/review.md +31 -31
  57. package/payload/playbook/routing.md +19 -19
  58. package/payload/playbook/rules-card.md +17 -16
  59. package/payload/playbook/ship.md +35 -35
  60. package/payload/playbook/steer.md +21 -21
  61. package/payload/playbook/verify.md +42 -42
  62. package/payload/templates/change-deep.md +29 -29
  63. package/payload/templates/change-standard.md +28 -28
  64. package/payload/templates/change-vibe.md +19 -19
  65. package/payload/templates/config.yaml +8 -8
  66. package/payload/templates/constitution.md +14 -14
  67. package/payload/templates/contract-evals.md +9 -9
  68. package/payload/templates/contract-tests.md +9 -9
  69. package/payload/templates/harness.md +34 -34
  70. package/payload/templates/spec.md +14 -14
  71. package/payload/templates/steering.md +9 -9
  72. package/scripts/validate.mjs +47 -47
@@ -1,35 +1,35 @@
1
- # /sdlc:ship <id> — merge, archive, learn, digest
2
-
3
- Precondition: `status: verified` (+ review passed when it ran).
4
-
5
- 1. **Policy check** (`sdlc/harness.md § Autonomy policy`): if this change is NOT
6
- auto-shippable (deep/hard-floor), show the human a 5-line summary (why, delta
7
- heads, verify result, cost so far) and wait for approval. Otherwise proceed.
8
- 2. Open the gate and archive mechanically:
9
- `node sdlc/.hooks/journal.mjs open-ship <id>`
10
- `npx @warnyin/sdlc archive <id>`
11
- (validates --strict, merges every Delta into `sdlc/specs/`, promotes evals,
12
- stamps `status: shipped`, moves the folder to `changes/archive/<date>-<id>/`).
13
- Any `⚠ MODIFIED Requirement …` line means the new body carried away a scenario
14
- the spec still promised: confirm it was intended and name it in the digest.
15
- 3. **Learn** — delegate to `sdlc-learner` (cheap) with the archived change.md +
16
- its journal.ndjson. It proposes ≤3 items: add-rule (with evidence pointer) /
17
- expire-or-demote (rule or steering that never fired) / harness tweak.
18
- Apply reductions and demotions immediately (they always save tokens).
19
- Additions to always-loaded context are NOT applied — list them in the digest.
20
- 4. **Digest** — write `sdlc/changes/archive/<date>-<id>/digest.md` (≤15 lines):
21
- what shipped, spec deltas merged, assumptions made, verify rounds, tokens/cost
22
- (from journal `session` events), learner proposals awaiting the human.
23
- When any verify or review note carries `mode=solo`, the digest SHALL say which
24
- outcomes were self-produced. A reader months from now cannot otherwise tell a
25
- panel's verdict from the author's own.
26
- When any `escalation` event carries `preauth=yes`, the digest SHALL list those
27
- pre-authorized escalations by condition — the points where a human would normally
28
- have stood and, this run, did not.
29
- 5. Close the gate: `node sdlc/.hooks/journal.mjs close`. Tell the user in one
30
- line: shipped + where the digest is.
31
-
32
- The digest is the async human touchpoint — reviewable and revertible later.
33
-
34
- `--auto`: do this stage, then continue to ship under `auto.md`'s unattended
35
- mode — gather, confirm once, run. The stage still does its own work first.
1
+ # /sdlc:ship <id> — merge, archive, learn, digest
2
+
3
+ Precondition: `status: verified` (+ review passed when it ran).
4
+
5
+ 1. **Policy check** (`sdlc/harness.md § Autonomy policy`): if this change is NOT
6
+ auto-shippable (deep/hard-floor), show the human a 5-line summary (why, delta
7
+ heads, verify result, cost so far) and wait for approval. Otherwise proceed.
8
+ 2. Open the gate and archive mechanically:
9
+ `node sdlc/.hooks/journal.mjs open-ship <id>`
10
+ `npx @warnyin/sdlc archive <id>`
11
+ (validates --strict, merges every Delta into `sdlc/specs/`, promotes evals,
12
+ stamps `status: shipped`, moves the folder to `changes/archive/<date>-<id>/`).
13
+ Any `⚠ MODIFIED Requirement …` line means the new body carried away a scenario
14
+ the spec still promised: confirm it was intended and name it in the digest.
15
+ 3. **Learn** — delegate to `sdlc-learner` (cheap) with the archived change.md +
16
+ its journal.ndjson. It proposes ≤3 items: add-rule (with evidence pointer) /
17
+ expire-or-demote (rule or steering that never fired) / harness tweak.
18
+ Apply reductions and demotions immediately (they always save tokens).
19
+ Additions to always-loaded context are NOT applied — list them in the digest.
20
+ 4. **Digest** — write `sdlc/changes/archive/<date>-<id>/digest.md` (≤15 lines):
21
+ what shipped, spec deltas merged, assumptions made, verify rounds, tokens/cost
22
+ (from journal `session` events), learner proposals awaiting the human.
23
+ When any verify or review note carries `mode=solo`, the digest SHALL say which
24
+ outcomes were self-produced. A reader months from now cannot otherwise tell a
25
+ panel's verdict from the author's own.
26
+ When any `escalation` event carries `preauth=yes`, the digest SHALL list those
27
+ pre-authorized escalations by condition — the points where a human would normally
28
+ have stood and, this run, did not.
29
+ 5. Close the gate: `node sdlc/.hooks/journal.mjs close`. Tell the user in one
30
+ line: shipped + where the digest is.
31
+
32
+ The digest is the async human touchpoint — reviewable and revertible later.
33
+
34
+ `--auto`: do this stage, then continue to ship under `auto.md`'s unattended
35
+ mode — gather, confirm once, run. The stage still does its own work first.
@@ -1,21 +1,21 @@
1
- # /sdlc:steer — manage the static/dynamic boundary
2
-
3
- The ONLY sanctioned way to edit always-loaded context.
4
-
5
- 1. `node sdlc/.hooks/journal.mjs open-steer`
6
- 2. Apply the requested edits:
7
- - New knowledge → a steering file with the LEAST resident inclusion mode that
8
- works: `agent` > `manual` > `paths` > `always` (always needs justification).
9
- - Constitution edits: SHALL/SHALL NOT lines only. If the always-budget
10
- (60 lines) would overflow, something must be demoted or deleted first —
11
- name the displaced line explicitly.
12
- 3. Hygiene pass (do it every time, it is 30 seconds):
13
- - `/sdlc:observe` data — steering never pointer-hit across the last N changes
14
- → propose demote (`paths`→`manual`) or delete.
15
- - Rules with no guard/journal evidence of ever mattering → propose expiry.
16
- - Apply what the user agrees to (reductions may be applied without asking).
17
- 4. `npx @warnyin/sdlc validate` — the always-budget check must be green.
18
- 5. `node sdlc/.hooks/journal.mjs close`
19
-
20
- Learning direction is one-way by design: content flows from always → paths →
21
- manual → deleted as evidence accumulates. Getting leaner is the success metric.
1
+ # /sdlc:steer — manage the static/dynamic boundary
2
+
3
+ The ONLY sanctioned way to edit always-loaded context.
4
+
5
+ 1. `node sdlc/.hooks/journal.mjs open-steer`
6
+ 2. Apply the requested edits:
7
+ - New knowledge → a steering file with the LEAST resident inclusion mode that
8
+ works: `agent` > `manual` > `paths` > `always` (always needs justification).
9
+ - Constitution edits: SHALL/SHALL NOT lines only. If the always-budget
10
+ (60 lines) would overflow, something must be demoted or deleted first —
11
+ name the displaced line explicitly.
12
+ 3. Hygiene pass (do it every time, it is 30 seconds):
13
+ - `/sdlc:observe` data — steering never pointer-hit across the last N changes
14
+ → propose demote (`paths`→`manual`) or delete.
15
+ - Rules with no guard/journal evidence of ever mattering → propose expiry.
16
+ - Apply what the user agrees to (reductions may be applied without asking).
17
+ 4. `npx @warnyin/sdlc validate` — the always-budget check must be green.
18
+ 5. `node sdlc/.hooks/journal.mjs close`
19
+
20
+ Learning direction is one-way by design: content flows from always → paths →
21
+ manual → deleted as evidence accumulates. Getting leaner is the success metric.
@@ -1,42 +1,42 @@
1
- # /sdlc:verify <id> — the feedback loop
2
-
3
- Two halves, both must pass. Verification is against the CONTRACT, not vibes.
4
-
5
- 1. **Tests (deterministic)**: run the full test command from `sdlc/harness.md`.
6
- Every row of `contract/tests.md` must be covered by a passing test.
7
- 2. **Evals (non-deterministic)** — when `contract/evals.md` exists: delegate to
8
- the `sdlc-evaluator` agent (cheap) with the rubric + the diff + the task log;
9
- it returns a score per rubric line. Pass bar is written in the file.
10
- If the evaluator cannot run — subagents unavailable or disallowed in this
11
- session — score in the main loop instead and record that. A panel that could
12
- not run is a fact to write down, never a reason to stop the pipeline; but a
13
- run that judged its own work is weaker evidence and must not read as if a
14
- panel had agreed.
15
-
16
- 3. **Lens bars** — a non-empty `lenses`, whether or not `evals.md` exists: every lens bar
17
- must appear as a `tests.md` row or an `evals.md` line; a lens with its bars in neither
18
- is a contract gap and fails verify.
19
-
20
- On failure:
21
- - Cluster failures by root cause (one line each) and append the cluster note to
22
- the change's `## Tasks` area as unchecked fix tasks.
23
- - `node sdlc/.hooks/journal.mjs note verify result=fail round=<n> mode=<panel|solo>`
24
- - Route back to /sdlc:build. Maximum 3 rounds total; on the 4th failure STOP and
25
- escalate to the human with the cluster history (Autonomy policy condition).
26
- - Never lower the bar: do not edit tests/evals to pass unless the contract
27
- itself was wrong — changing the contract reopens the adversarial check.
28
-
29
- On pass: set `status: verified`,
30
- `node sdlc/.hooks/journal.mjs note verify result=pass round=<n> mode=<panel|solo>`.
31
-
32
- `mode=panel` only when independent agents produced the judgment; `mode=solo` when
33
- the main loop judged its own work. Every verify note carries it, pass or fail —
34
- `/sdlc:observe` reports a change as self-judged from this field, and omitting it
35
- leaves the record silently indistinguishable from an independent one.
36
-
37
- Next: review signals present (deep tier, security-touching diff, >10 files,
38
- a non-empty `lenses`)
39
- → /sdlc:review; otherwise → /sdlc:ship.
40
-
41
- `--auto`: do this stage, then continue to ship under `auto.md`'s unattended
42
- mode — gather, confirm once, run. The stage still does its own work first.
1
+ # /sdlc:verify <id> — the feedback loop
2
+
3
+ Two halves, both must pass. Verification is against the CONTRACT, not vibes.
4
+
5
+ 1. **Tests (deterministic)**: run the full test command from `sdlc/harness.md`.
6
+ Every row of `contract/tests.md` must be covered by a passing test.
7
+ 2. **Evals (non-deterministic)** — when `contract/evals.md` exists: delegate to
8
+ the `sdlc-evaluator` agent (cheap) with the rubric + the diff + the task log;
9
+ it returns a score per rubric line. Pass bar is written in the file.
10
+ If the evaluator cannot run — subagents unavailable or disallowed in this
11
+ session — score in the main loop instead and record that. A panel that could
12
+ not run is a fact to write down, never a reason to stop the pipeline; but a
13
+ run that judged its own work is weaker evidence and must not read as if a
14
+ panel had agreed.
15
+
16
+ 3. **Lens bars** — a non-empty `lenses`, whether or not `evals.md` exists: every lens bar
17
+ must appear as a `tests.md` row or an `evals.md` line; a lens with its bars in neither
18
+ is a contract gap and fails verify.
19
+
20
+ On failure:
21
+ - Cluster failures by root cause (one line each) and append the cluster note to
22
+ the change's `## Tasks` area as unchecked fix tasks.
23
+ - `node sdlc/.hooks/journal.mjs note verify result=fail round=<n> mode=<panel|solo>`
24
+ - Route back to /sdlc:build. Maximum 3 rounds total; on the 4th failure STOP and
25
+ escalate to the human with the cluster history (Autonomy policy condition).
26
+ - Never lower the bar: do not edit tests/evals to pass unless the contract
27
+ itself was wrong — changing the contract reopens the adversarial check.
28
+
29
+ On pass: set `status: verified`,
30
+ `node sdlc/.hooks/journal.mjs note verify result=pass round=<n> mode=<panel|solo>`.
31
+
32
+ `mode=panel` only when independent agents produced the judgment; `mode=solo` when
33
+ the main loop judged its own work. Every verify note carries it, pass or fail —
34
+ `/sdlc:observe` reports a change as self-judged from this field, and omitting it
35
+ leaves the record silently indistinguishable from an independent one.
36
+
37
+ Next: review signals present (deep tier, security-touching diff, >10 files,
38
+ a non-empty `lenses`)
39
+ → /sdlc:review; otherwise → /sdlc:ship.
40
+
41
+ `--auto`: do this stage, then continue to ship under `auto.md`'s unattended
42
+ mode — gather, confirm once, run. The stage still does its own work first.
@@ -1,29 +1,29 @@
1
- ---
2
- id: <kebab-id>
3
- tier: deep
4
- status: new
5
- ---
6
- # Change: <title>
7
- <!-- cap:150 · deep tier: hard-floor surface (security/payments/data-loss/irreversible) or new/multi-capability delta. Human gates apply per Autonomy policy. -->
8
-
9
- ## Why (≤5 lines)
10
- <problem + outcome. No solutioning.>
11
-
12
- ## Assumptions
13
- - <assumption made instead of asking, with the reason it is safe>
14
-
15
- ## Delta: <capability>
16
-
17
- ### ADDED Requirement: <name>
18
- The system SHALL <behavior>.
19
-
20
- #### Scenario: <name>
21
- - WHEN <condition>
22
- - THEN <observable outcome>
23
-
24
- ## Design
25
- <!-- decisions & trade-offs only — never restate the delta. Escalate irreversible decisions per Autonomy policy. -->
26
- - decision: <what> · alternatives: <a/b> · because: <why>
27
-
28
- ## Tasks
29
- - [ ] T1 <task> [P] [tier:balanced]
1
+ ---
2
+ id: <kebab-id>
3
+ tier: deep
4
+ status: new
5
+ ---
6
+ # Change: <title>
7
+ <!-- cap:150 · deep tier: hard-floor surface (security/payments/data-loss/irreversible) or new/multi-capability delta. Human gates apply per Autonomy policy. -->
8
+
9
+ ## Why (≤5 lines)
10
+ <problem + outcome. No solutioning.>
11
+
12
+ ## Assumptions
13
+ - <assumption made instead of asking, with the reason it is safe>
14
+
15
+ ## Delta: <capability>
16
+
17
+ ### ADDED Requirement: <name>
18
+ The system SHALL <behavior>.
19
+
20
+ #### Scenario: <name>
21
+ - WHEN <condition>
22
+ - THEN <observable outcome>
23
+
24
+ ## Design
25
+ <!-- decisions & trade-offs only — never restate the delta. Escalate irreversible decisions per Autonomy policy. -->
26
+ - decision: <what> · alternatives: <a/b> · because: <why>
27
+
28
+ ## Tasks
29
+ - [ ] T1 <task> [P] [tier:balanced]
@@ -1,28 +1,28 @@
1
- ---
2
- id: <kebab-id>
3
- tier: standard
4
- status: new
5
- ---
6
- # Change: <title>
7
- <!-- cap:100 · standard tier. Delta sections ARE the spec change — merged mechanically at ship, never re-narrated. -->
8
-
9
- ## Why (≤5 lines)
10
- <problem + outcome. No solutioning.>
11
-
12
- ## Assumptions
13
- - <assumption made instead of asking, with the reason it is safe — delete section if none>
14
- <!-- unresolved ambiguity the agent cannot assume → [NEEDS CLARIFICATION: question] inline where it bites -->
15
-
16
- ## Delta: <capability>
17
-
18
- ### ADDED Requirement: <name>
19
- The system SHALL <behavior>.
20
-
21
- #### Scenario: <name>
22
- - WHEN <condition>
23
- - THEN <observable outcome>
24
-
25
- ## Tasks
26
- <!-- [P] = parallelizable wave · [tier:cheap|balanced|deepest] per the routing table in harness.md -->
27
- - [ ] T1 <task> [P] [tier:balanced]
28
- - [ ] T2 <task> [tier:cheap]
1
+ ---
2
+ id: <kebab-id>
3
+ tier: standard
4
+ status: new
5
+ ---
6
+ # Change: <title>
7
+ <!-- cap:100 · standard tier. Delta sections ARE the spec change — merged mechanically at ship, never re-narrated. -->
8
+
9
+ ## Why (≤5 lines)
10
+ <problem + outcome. No solutioning.>
11
+
12
+ ## Assumptions
13
+ - <assumption made instead of asking, with the reason it is safe — delete section if none>
14
+ <!-- unresolved ambiguity the agent cannot assume → [NEEDS CLARIFICATION: question] inline where it bites -->
15
+
16
+ ## Delta: <capability>
17
+
18
+ ### ADDED Requirement: <name>
19
+ The system SHALL <behavior>.
20
+
21
+ #### Scenario: <name>
22
+ - WHEN <condition>
23
+ - THEN <observable outcome>
24
+
25
+ ## Tasks
26
+ <!-- [P] = parallelizable wave · [tier:cheap|balanced|deepest] per the routing table in harness.md -->
27
+ - [ ] T1 <task> [P] [tier:balanced]
28
+ - [ ] T2 <task> [tier:cheap]
@@ -1,19 +1,19 @@
1
- ---
2
- id: <kebab-id>
3
- tier: vibe
4
- status: new
5
- ---
6
- # Change: <title>
7
- <!-- cap:40 · vibe tier = one file, no contract. Reversible, no behavior-contract change, ≤2 files. -->
8
-
9
- ## Why (≤3 lines)
10
- <problem + outcome>
11
-
12
- ## Assumptions
13
- - <assumption the agent made instead of asking — delete section if none>
14
-
15
- ## Tasks
16
- - [ ] T1 <task>
17
-
18
- ## Receipt (filled at ship)
19
- - files touched: · test result: · cost:
1
+ ---
2
+ id: <kebab-id>
3
+ tier: vibe
4
+ status: new
5
+ ---
6
+ # Change: <title>
7
+ <!-- cap:40 · vibe tier = one file, no contract. Reversible, no behavior-contract change, ≤2 files. -->
8
+
9
+ ## Why (≤3 lines)
10
+ <problem + outcome>
11
+
12
+ ## Assumptions
13
+ - <assumption the agent made instead of asking — delete section if none>
14
+
15
+ ## Tasks
16
+ - [ ] T1 <task>
17
+
18
+ ## Receipt (filled at ship)
19
+ - files touched: · test result: · cost:
@@ -1,8 +1,8 @@
1
- # @warnyin/sdlc project config — machine-read; keep minimal.
2
- language: en
3
- tools: [] # filled by `warnyin-sdlc init`
4
- # Optional price table for cost reporting (USD per 1M tokens).
5
- # prices:
6
- # claude-sonnet-5: { input: 3, output: 15, cacheRead: 0.3 }
7
- # Optional per-artifact cap overrides (see lib/caps.mjs for defaults).
8
- # caps: {}
1
+ # @warnyin/sdlc project config — machine-read; keep minimal.
2
+ language: en
3
+ tools: [] # filled by `warnyin-sdlc init`
4
+ # Optional price table for cost reporting (USD per 1M tokens).
5
+ # prices:
6
+ # claude-sonnet-5: { input: 3, output: 15, cacheRead: 0.3 }
7
+ # Optional per-artifact cap overrides (see lib/caps.mjs for defaults).
8
+ # caps: {}
@@ -1,14 +1,14 @@
1
- # Constitution — <project>
2
- <!-- cap:30 · ALWAYS LOADED: every line here costs tokens in EVERY turn. Prove residency or move it to steering. -->
3
-
4
- ## Stack (facts only, max 3 lines)
5
- - <runtime / framework / db — one line each>
6
-
7
- ## Hard rules (SHALL / SHALL NOT only — no advice)
8
- - The agent SHALL NOT edit `sdlc/specs/**` or `sdlc/changes/archive/**` outside `/sdlc:ship`.
9
- - The agent SHALL record every assumption in the change's `## Assumptions` before acting on it.
10
- - <add a rule each time the agent misbehaves; each new rule must displace an old one once the always-budget (60 lines) is full>
11
-
12
- ## Workflow
13
- - Changes flow: new → [design] → contract → build → verify → [review] → ship.
14
- - Tier by stakes: vibe | standard | deep — triage table lives in `sdlc/harness.md`.
1
+ # Constitution — <project>
2
+ <!-- cap:30 · ALWAYS LOADED: every line here costs tokens in EVERY turn. Prove residency or move it to steering. -->
3
+
4
+ ## Stack (facts only, max 3 lines)
5
+ - <runtime / framework / db — one line each>
6
+
7
+ ## Hard rules (SHALL / SHALL NOT only — no advice)
8
+ - The agent SHALL NOT edit `sdlc/specs/**` or `sdlc/changes/archive/**` outside `/sdlc:ship`.
9
+ - The agent SHALL record every assumption in the change's `## Assumptions` before acting on it.
10
+ - <add a rule each time the agent misbehaves; each new rule must displace an old one once the always-budget (60 lines) is full>
11
+
12
+ ## Workflow
13
+ - Changes flow: new → [design] → contract → build → verify → [review] → ship.
14
+ - Tier by stakes: vibe | standard | deep — triage table lives in `sdlc/harness.md`.
@@ -1,9 +1,9 @@
1
- # Eval contract — <change-id>
2
- <!-- cap:40 · verifies the non-deterministic half: trajectory + quality. Scored by sdlc-evaluator (cheap tier). -->
3
-
4
- ## Rubric (score 1–5 each)
5
- - Trajectory: did the agent read the contract before writing code? run tests before claiming done?
6
- - Quality: <change-specific bars>
7
-
8
- ## Pass bar
9
- - all scores ≥ 4 · failures route back to `/sdlc:build` with a cluster note
1
+ # Eval contract — <change-id>
2
+ <!-- cap:40 · verifies the non-deterministic half: trajectory + quality. Scored by sdlc-evaluator (cheap tier). -->
3
+
4
+ ## Rubric (score 1–5 each)
5
+ - Trajectory: did the agent read the contract before writing code? run tests before claiming done?
6
+ - Quality: <change-specific bars>
7
+
8
+ ## Pass bar
9
+ - all scores ≥ 4 · failures route back to `/sdlc:build` with a cluster note
@@ -1,9 +1,9 @@
1
- # Test contract — <change-id>
2
- <!-- cap:60 · written BEFORE code. This is the deterministic half of the contract with the AI; failing tests are generated from this table. -->
3
-
4
- | # | Given / When / Then | Kind (unit/int/e2e) | Maps to requirement |
5
- |---|---|---|---|
6
- | 1 | <g/w/t> | unit | <requirement name> |
7
-
8
- ## Out of scope (explicitly untested + why)
9
- - <item — reason>
1
+ # Test contract — <change-id>
2
+ <!-- cap:60 · written BEFORE code. This is the deterministic half of the contract with the AI; failing tests are generated from this table. -->
3
+
4
+ | # | Given / When / Then | Kind (unit/int/e2e) | Maps to requirement |
5
+ |---|---|---|---|
6
+ | 1 | <g/w/t> | unit | <requirement name> |
7
+
8
+ ## Out of scope (explicitly untested + why)
9
+ - <item — reason>
@@ -1,34 +1,34 @@
1
- # Harness — <project>
2
- <!-- cap:60 · CONFIGURE-phase output; reviewed like code. Registry + tables only — doctrine lives in sdlc/.playbook/. -->
3
-
4
- ## Tools & MCP (what the agent may call)
5
- | Tool/MCP | Scope | Notes |
6
- |---|---|---|
7
- | <tool> | <what it may touch> | <when to use> |
8
-
9
- ## Sandbox & execution
10
- - test command: `<cmd>`
11
- - sandbox notes: <where code runs, what it cannot reach>
12
-
13
- ## Guardrails (mirror of installed hooks — deterministic, the agent cannot skip them)
14
- - `sdlc/specs/**` and archive are write-locked outside ship.
15
- - Artifact line caps validated on every write.
16
- - Session token/cost journaled per change.
17
-
18
- ## Model routing (generic tiers — the harness adapter maps them to real models)
19
- | Task kind | Tier |
20
- |---|---|
21
- | requirements, architecture, deep design | deepest |
22
- | standard implementation | balanced |
23
- | test generation, review passes, mechanical/scaffold, eval judging, learning | cheap |
24
-
25
- ## Tier triage (stakes → tier)
26
- - vibe: reversible, no behavior-contract change, ≤2 files.
27
- - deep: security / payments / data-loss / irreversible surface, new capability, or multi-capability delta.
28
- - otherwise: standard.
29
-
30
- ## Autonomy policy
31
- - auto-ship: vibe, standard.
32
- - escalate to human: hard-floor (security, payments, data-loss, irreversible), verify failed > 3 rounds,
33
- token budget exceeded, information the agent cannot obtain or safely assume,
34
- installing a skill or agent (suggest only; see `.playbook/lenses.md`).
1
+ # Harness — <project>
2
+ <!-- cap:60 · CONFIGURE-phase output; reviewed like code. Registry + tables only — doctrine lives in sdlc/.playbook/. -->
3
+
4
+ ## Tools & MCP (what the agent may call)
5
+ | Tool/MCP | Scope | Notes |
6
+ |---|---|---|
7
+ | <tool> | <what it may touch> | <when to use> |
8
+
9
+ ## Sandbox & execution
10
+ - test command: `<cmd>`
11
+ - sandbox notes: <where code runs, what it cannot reach>
12
+
13
+ ## Guardrails (mirror of installed hooks — deterministic, the agent cannot skip them)
14
+ - `sdlc/specs/**` and archive are write-locked outside ship.
15
+ - Artifact line caps validated on every write.
16
+ - Session token/cost journaled per change.
17
+
18
+ ## Model routing (generic tiers — the harness adapter maps them to real models)
19
+ | Task kind | Tier |
20
+ |---|---|
21
+ | requirements, architecture, deep design | deepest |
22
+ | standard implementation | balanced |
23
+ | test generation, review passes, mechanical/scaffold, eval judging, learning | cheap |
24
+
25
+ ## Tier triage (stakes → tier)
26
+ - vibe: reversible, no behavior-contract change, ≤2 files.
27
+ - deep: security / payments / data-loss / irreversible surface, new capability, or multi-capability delta.
28
+ - otherwise: standard.
29
+
30
+ ## Autonomy policy
31
+ - auto-ship: vibe, standard.
32
+ - escalate to human: hard-floor (security, payments, data-loss, irreversible), verify failed > 3 rounds,
33
+ token budget exceeded, information the agent cannot obtain or safely assume,
34
+ installing a skill or agent (suggest only; see `.playbook/lenses.md`).
@@ -1,14 +1,14 @@
1
- # Spec: <capability>
2
- <!-- cap:150 (soft) · living spec — observable behavior only, no implementation detail. Split the capability beyond the cap. -->
3
-
4
- ## Purpose
5
- <!-- one or two lines; commands grep this header first (progressive disclosure) -->
6
-
7
- ## Requirements
8
-
9
- ### Requirement: <name>
10
- The system SHALL <behavior>.
11
-
12
- #### Scenario: <name>
13
- - WHEN <condition>
14
- - THEN <observable outcome>
1
+ # Spec: <capability>
2
+ <!-- cap:150 (soft) · living spec — observable behavior only, no implementation detail. Split the capability beyond the cap. -->
3
+
4
+ ## Purpose
5
+ <!-- one or two lines; commands grep this header first (progressive disclosure) -->
6
+
7
+ ## Requirements
8
+
9
+ ### Requirement: <name>
10
+ The system SHALL <behavior>.
11
+
12
+ #### Scenario: <name>
13
+ - WHEN <condition>
14
+ - THEN <observable outcome>
@@ -1,9 +1,9 @@
1
- ---
2
- name: <topic>
3
- inclusion: paths # always | paths | manual | agent
4
- pathMatch: ["src/<area>/**"]
5
- ---
6
- # Steering: <topic>
7
- <!-- cap:40 · loaded only when pathMatch hits or explicitly requested. Dense, scoped conventions — no generalities. -->
8
-
9
- <conventions, gotchas, and reference patterns for this scope>
1
+ ---
2
+ name: <topic>
3
+ inclusion: paths # always | paths | manual | agent
4
+ pathMatch: ["src/<area>/**"]
5
+ ---
6
+ # Steering: <topic>
7
+ <!-- cap:40 · loaded only when pathMatch hits or explicitly requested. Dense, scoped conventions — no generalities. -->
8
+
9
+ <conventions, gotchas, and reference patterns for this scope>