pantheon-opencode 1.7.0-beta.1 → 2.0.0-beta.1
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.
- package/AGENTS.global.md +2 -2
- package/AGENTS.md +3 -3
- package/CHANGELOG.md +86 -1
- package/README.md +90 -45
- package/README.pt-BR.md +19 -19
- package/bin/pantheon-init.mjs +11 -48
- package/package.json +9 -13
- package/plugin.json +2 -2
- package/pyproject.toml +1 -1
- package/scripts/build-agents-md.mjs +7 -6
- package/scripts/{check-plugin-v2-coverage.mjs → check-plugin-coverage.mjs} +7 -7
- package/scripts/{decision-router-v2-budget.mjs → decision-router-budget-gates.mjs} +1 -1
- package/scripts/{decision-router-v2-egress.mjs → decision-router-egress.mjs} +1 -1
- package/scripts/decision-router.mjs +2 -2
- package/scripts/docs-media.mjs +1 -1
- package/scripts/doctor.mjs +225 -136
- package/scripts/install/behavior-overrides.bundle.mjs +123 -0
- package/scripts/install/{v2-capabilities.mjs → capabilities.mjs} +7 -7
- package/scripts/install/config-migration.mjs +60 -253
- package/scripts/install/file-inventory.mjs +6 -4
- package/scripts/install/interactive.mjs +9 -2
- package/scripts/install/opencode-state.mjs +13 -12
- package/scripts/install/opencode-version.mjs +97 -162
- package/scripts/install/opencode.mjs +271 -428
- package/scripts/install/plan.mjs +1 -9
- package/scripts/install/plugin.mjs +19 -35
- package/scripts/install/shared.mjs +40 -1
- package/scripts/package-evidence.mjs +1 -4
- package/scripts/probe-pantheon-cost.mjs +1 -1
- package/scripts/reconcile.mjs +5 -7
- package/scripts/release-gates.mjs +38 -58
- package/scripts/setup-v2-config.mjs +10 -8
- package/scripts/setup.mjs +12 -22
- package/scripts/sync-tui.mjs +9 -17
- package/scripts/{test-opencode-v2-sandbox.sh → test-opencode-sandbox.sh} +9 -9
- package/scripts/update-skills-lock.mjs +24 -12
- package/scripts/update.mjs +17 -1
- package/scripts/validate-routing.mjs +79 -4
- package/scripts/validate-sandbox-install.mjs +3 -3
- package/skills-lock.json +25 -8
- package/src/agents/zeus.md +1 -0
- package/src/pantheon/behavior-overrides.ts +191 -0
- package/src/pantheon/{v2-bounded-read.ts → bounded-read.ts} +1 -1
- package/src/pantheon/cost-command.ts +1 -1
- package/src/pantheon/{v2-decision-contracts.ts → decision-contracts.ts} +3 -3
- package/src/pantheon/{v2-decision-points → decision-points}/p1-agent.ts +3 -3
- package/src/pantheon/{v2-decision-points → decision-points}/p10-claims.ts +2 -2
- package/src/pantheon/{v2-decision-points → decision-points}/p11-catalog.ts +2 -2
- package/src/pantheon/{v2-decision-points → decision-points}/p2-skill.ts +3 -3
- package/src/pantheon/{v2-decision-points → decision-points}/p3-complexity.ts +6 -6
- package/src/pantheon/{v2-decision-points → decision-points}/p4-memory.ts +2 -2
- package/src/pantheon/{v2-decision-points → decision-points}/p5-themis-verify.ts +2 -2
- package/src/pantheon/{v2-decision-points → decision-points}/p6-council.ts +3 -3
- package/src/pantheon/{v2-decision-points → decision-points}/p7-loop.ts +5 -5
- package/src/pantheon/{v2-decision-points → decision-points}/p8-guardrail.ts +2 -2
- package/src/pantheon/{v2-decision-points → decision-points}/p9-memory-compaction.ts +2 -2
- package/src/pantheon/{v2-decision-points.ts → decision-points.ts} +5 -5
- package/src/pantheon/{v2-decision-policy.ts → decision-policy.ts} +2 -2
- package/src/pantheon/{v2-decision-prompt-surface.ts → decision-prompt-surface.ts} +3 -3
- package/src/pantheon/{v2-decision-registry.ts → decision-registry.ts} +10 -13
- package/src/pantheon/{v2-decision-resource-ids.ts → decision-resource-ids.ts} +1 -1
- package/src/pantheon/{v2-decision-shadow.ts → decision-shadow.ts} +1 -1
- package/src/pantheon/{v2-decision-telemetry.ts → decision-telemetry.ts} +1 -1
- package/src/pantheon/{v2-execute-after.ts → execute-after.ts} +4 -4
- package/src/pantheon/goal-store.ts +7 -0
- package/src/pantheon/{v2-hook-engine.ts → hook-engine.ts} +3 -3
- package/src/pantheon/jev-decision-tool.ts +3 -3
- package/src/pantheon/jev-decisions.ts +12 -6
- package/src/pantheon/jev-escalation.ts +6 -6
- package/src/pantheon/jev-quality-score.ts +2 -2
- package/src/pantheon/jev-rank-verify.ts +2 -2
- package/src/pantheon/jev-risk-gate.ts +4 -4
- package/src/pantheon/local-decision-agents.ts +1 -1
- package/src/pantheon/local-decision-router.ts +1 -1
- package/src/pantheon/local-decision-tool.ts +2 -2
- package/src/pantheon/local-orchestration.ts +1 -1
- package/src/pantheon/model-command.ts +1 -1
- package/src/pantheon/{v2-vision.ts → native-vision.ts} +12 -15
- package/src/pantheon/pantheon-decision-catalog.ts +12 -0
- package/src/pantheon/{v2-shell-hooks.ts → shell-hooks.ts} +4 -4
- package/src/pantheon/{v2-tools.ts → tools.ts} +4 -4
- package/src/pantheon/{v2-unsupported.d.mts → unsupported.d.mts} +3 -3
- package/src/pantheon/{v2-unsupported.mjs → unsupported.mjs} +8 -8
- package/src/pantheon/update-notifier.ts +101 -0
- package/src/pantheon/vision-local.ts +12 -3
- package/src/pantheon/vision.ts +120 -2
- package/src/{plugin-v2 → plugin}/index.ts +7 -7
- package/src/plugin.ts +2875 -500
- package/src/plugins/hook-runner.ts +3 -3
- package/src/plugins/tui/dist/server.d.ts +23 -9
- package/src/plugins/tui/dist/server.d.ts.map +1 -1
- package/src/plugins/tui/dist/server.js +25 -12
- package/src/plugins/tui/dist/server.js.map +1 -1
- package/src/plugins/tui/dist/tui.d.ts +1023 -494
- package/src/plugins/tui/dist/tui.d.ts.map +1 -1
- package/src/plugins/tui/dist/tui.js +4841 -2066
- package/src/plugins/tui/dist/tui.js.map +1 -1
- package/src/plugins/tui/{index-v2.ts → index.ts} +1 -1
- package/src/plugins/tui/package-lock.json +2 -42
- package/src/plugins/tui/package.json +4 -10
- package/src/plugins/tui/src/cost-data.ts +1 -1
- package/src/plugins/tui/src/index.tsx +4222 -2822
- package/src/plugins/tui/src/pantheon-settings-state.ts +111 -0
- package/src/plugins/tui/src/pantheon-settings.tsx +236 -0
- package/src/plugins/tui/src/server.ts +23 -10
- package/src/plugins/tui/src/update-notice.ts +86 -0
- package/src/plugins/tui/tsconfig.typecheck.json +1 -1
- package/src/plugins/tui/tsdown.config.ts +3 -5
- package/src/plugins/tui/tui.ts +1 -2
- package/src/routing.yml +36 -0
- package/src/skills/README.md +16 -1
- package/src/skills/artifact-management/SKILL.md +10 -67
- package/src/skills/artifact-management/references/artifact-templates.md +71 -0
- package/src/skills/auto-continue/SKILL.md +10 -0
- package/src/skills/code-review-checklist/SKILL.md +11 -44
- package/src/skills/code-review-checklist/references/review-checklists.md +54 -0
- package/src/skills/instruction-forge/SKILL.md +57 -0
- package/src/skills/instruction-forge/references/agents-md-doctrine.md +50 -0
- package/src/skills/instruction-forge/references/eval-doctrine.md +37 -0
- package/src/skills/instruction-forge/references/skill-doctrine.md +62 -0
- package/src/skills/memory-bank/SKILL.md +11 -90
- package/src/skills/memory-bank/references/anti-patterns.md +33 -0
- package/src/skills/memory-bank/references/product-adoption.md +19 -0
- package/src/skills/memory-bank/references/storage-model.md +21 -0
- package/src/skills/memory-bank/references/structure-and-frequency.md +44 -0
- package/src/skills/reflect/SKILL.md +1 -0
- package/src/skills/session-goal/SKILL.md +7 -54
- package/src/skills/session-goal/references/goal-format.md +26 -0
- package/src/skills/session-goal/references/session-reuse.md +44 -0
- package/src/skills/tdd-with-agents/SKILL.md +11 -101
- package/src/skills/tdd-with-agents/references/advanced-testing-patterns.md +58 -0
- package/src/skills/tdd-with-agents/references/testing-by-layer.md +60 -0
- package/src/skills/visual-review-pipeline/SKILL.md +11 -120
- package/src/skills/visual-review-pipeline/references/findings-schema.md +49 -0
- package/src/skills/visual-review-pipeline/references/workflow-steps.md +111 -0
- package/scripts/hooks/log-session-start.sh +0 -48
- package/scripts/hooks/validate-post-conditions.sh +0 -21
- package/scripts/hooks/validate-talos-scope.sh +0 -50
- package/scripts/hooks/validate-tool-safety.sh +0 -49
- package/src/pantheon/goal-loop.ts +0 -314
- package/src/pantheon/idle-continuation.ts +0 -52
- package/src/pantheon/todo-preserve.ts +0 -301
- package/src/pantheon/tool-ceiling.ts +0 -159
- package/src/pantheon/v2-bridge.ts +0 -153
- package/src/plugin-v2.ts +0 -2954
- package/src/plugins/pantheon-hooks.ts +0 -1433
- package/src/plugins/tui/dist/server-v2.d.ts +0 -30
- package/src/plugins/tui/dist/server-v2.d.ts.map +0 -1
- package/src/plugins/tui/dist/server-v2.js +0 -33
- package/src/plugins/tui/dist/server-v2.js.map +0 -1
- package/src/plugins/tui/dist/tui-v2.d.ts +0 -1039
- package/src/plugins/tui/dist/tui-v2.d.ts.map +0 -1
- package/src/plugins/tui/dist/tui-v2.js +0 -4597
- package/src/plugins/tui/dist/tui-v2.js.map +0 -1
- package/src/plugins/tui/src/index-v2.tsx +0 -4496
- package/src/plugins/tui/src/server-v2.ts +0 -28
- package/src/plugins/tui/tui-v2.ts +0 -2
- /package/src/pantheon/{v2-decision-budget.ts → decision-budget.ts} +0 -0
- /package/src/pantheon/{v2-decision-egress.ts → decision-egress.ts} +0 -0
- /package/src/pantheon/{v2-decision-events.ts → decision-events.ts} +0 -0
- /package/src/pantheon/{v2-decision-heuristic.ts → decision-heuristic.ts} +0 -0
- /package/src/pantheon/{v2-decision-premises.ts → decision-premises.ts} +0 -0
- /package/src/pantheon/{v2-decision-rank.ts → decision-rank.ts} +0 -0
- /package/src/pantheon/{v2-decision-verify.ts → decision-verify.ts} +0 -0
- /package/src/pantheon/{v2-delegation-enforce.ts → delegation-policy.ts} +0 -0
- /package/src/pantheon/{v2-model-overrides.ts → model-overrides.ts} +0 -0
- /package/src/pantheon/{v2-native-decisions.ts → native-decisions.ts} +0 -0
- /package/src/plugins/tui/src/{opencode-v2.d.ts → opencode.d.ts} +0 -0
|
@@ -9,29 +9,6 @@ Structured artifact trail for planned, multi-phase, or materially risky implemen
|
|
|
9
9
|
|
|
10
10
|
---
|
|
11
11
|
|
|
12
|
-
## Temp Folder
|
|
13
|
-
|
|
14
|
-
All ephemeral artifacts live in `.pantheon/memory-bank/.tmp/` — gitignored and wiped on sprint close.
|
|
15
|
-
|
|
16
|
-
```
|
|
17
|
-
.pantheon/memory-bank/
|
|
18
|
-
├── .tmp/ ← GITIGNORED — ephemeral artifacts
|
|
19
|
-
│ ├── PLAN-<feature>.md
|
|
20
|
-
│ ├── IMPL-phase1-hermes.md
|
|
21
|
-
│ ├── IMPL-phase1-aphrodite.md
|
|
22
|
-
│ └── REVIEW-<feature>.md
|
|
23
|
-
├── _notes/ ← permanent ADRs; Git tracking follows repository policy
|
|
24
|
-
│ └── ADR-<topic>.md
|
|
25
|
-
├── 01-active-context.md
|
|
26
|
-
└── 02-progress-log.md
|
|
27
|
-
```
|
|
28
|
-
|
|
29
|
-
Memory Bank files are not automatically versioned or shared across clones.
|
|
30
|
-
Check the repository's ignore/commit policy before relying on `_tasks/` records
|
|
31
|
-
or other files being tracked; `_tasks/` may be Git-ignored.
|
|
32
|
-
|
|
33
|
-
---
|
|
34
|
-
|
|
35
12
|
## Artifact Types
|
|
36
13
|
|
|
37
14
|
| Prefix | Location | Ephemeral? | Produced by |
|
|
@@ -65,50 +42,6 @@ or other files being tracked; `_tasks/` may be Git-ignored.
|
|
|
65
42
|
|
|
66
43
|
---
|
|
67
44
|
|
|
68
|
-
## Templates
|
|
69
|
-
|
|
70
|
-
### PLAN
|
|
71
|
-
```markdown
|
|
72
|
-
# PLAN-<feature>
|
|
73
|
-
**Date:** YYYY-MM-DD **Status:** Awaiting Approval
|
|
74
|
-
|
|
75
|
-
## Goal
|
|
76
|
-
[One sentence]
|
|
77
|
-
|
|
78
|
-
## Phases
|
|
79
|
-
1. Phase 1 — @hermes
|
|
80
|
-
2. Phase 2 — @aphrodite
|
|
81
|
-
|
|
82
|
-
## Risks
|
|
83
|
-
- [Risk]
|
|
84
|
-
```
|
|
85
|
-
|
|
86
|
-
### IMPL
|
|
87
|
-
```markdown
|
|
88
|
-
# IMPL-<phase>-<agent>
|
|
89
|
-
**Date:** YYYY-MM-DD **Status:** Awaiting Themis Review
|
|
90
|
-
|
|
91
|
-
## What Was Implemented
|
|
92
|
-
- [file] — [what changed]
|
|
93
|
-
|
|
94
|
-
## Tests
|
|
95
|
-
- ✅ X tests / Coverage: Y%
|
|
96
|
-
```
|
|
97
|
-
|
|
98
|
-
### REVIEW
|
|
99
|
-
```markdown
|
|
100
|
-
# REVIEW-<feature>
|
|
101
|
-
**Status:** APPROVED | NEEDS_REVISION | FAILED
|
|
102
|
-
|
|
103
|
-
## Verdict
|
|
104
|
-
[APPROVED | NEEDS_REVISION | FAILED]
|
|
105
|
-
|
|
106
|
-
## Issues
|
|
107
|
-
- CRITICAL: X / HIGH: Y / MEDIUM: Z
|
|
108
|
-
```
|
|
109
|
-
|
|
110
|
-
---
|
|
111
|
-
|
|
112
45
|
## Human Pause Points
|
|
113
46
|
|
|
114
47
|
1. **After a required PLAN** → pause only when a material scope/risk decision needs approval.
|
|
@@ -124,3 +57,13 @@ or other files being tracked; `_tasks/` may be Git-ignored.
|
|
|
124
57
|
@mnemosyne Clean tmp # Wipes .tmp/ without closing sprint
|
|
125
58
|
@mnemosyne List artifacts # Check what's in .tmp/
|
|
126
59
|
```
|
|
60
|
+
|
|
61
|
+
---
|
|
62
|
+
|
|
63
|
+
## Routing
|
|
64
|
+
|
|
65
|
+
| Action / trigger | Reference |
|
|
66
|
+
|---|---|
|
|
67
|
+
| Creating an artifact: the `.tmp/` layout or the PLAN/IMPL/REVIEW templates | [references/artifact-templates.md](references/artifact-templates.md) |
|
|
68
|
+
|
|
69
|
+
A missing reference interrupts the action — read nothing in its place, never improvise from memory.
|
|
@@ -0,0 +1,71 @@
|
|
|
1
|
+
# Artifact Management — Templates
|
|
2
|
+
|
|
3
|
+
Detail companion for `SKILL.md`. Read on demand when creating an artifact and
|
|
4
|
+
you need the `.tmp/` layout or a PLAN/IMPL/REVIEW template. Artifact types,
|
|
5
|
+
who-generates-what, proportional-use rules, pause points, and cleanup stay in
|
|
6
|
+
`SKILL.md`.
|
|
7
|
+
|
|
8
|
+
---
|
|
9
|
+
|
|
10
|
+
## Temp Folder
|
|
11
|
+
|
|
12
|
+
All ephemeral artifacts live in `.pantheon/memory-bank/.tmp/` — gitignored and wiped on sprint close.
|
|
13
|
+
|
|
14
|
+
```
|
|
15
|
+
.pantheon/memory-bank/
|
|
16
|
+
├── .tmp/ ← GITIGNORED — ephemeral artifacts
|
|
17
|
+
│ ├── PLAN-<feature>.md
|
|
18
|
+
│ ├── IMPL-phase1-hermes.md
|
|
19
|
+
│ ├── IMPL-phase1-aphrodite.md
|
|
20
|
+
│ └── REVIEW-<feature>.md
|
|
21
|
+
├── _notes/ ← permanent ADRs; Git tracking follows repository policy
|
|
22
|
+
│ └── ADR-<topic>.md
|
|
23
|
+
├── 01-active-context.md
|
|
24
|
+
└── 02-progress-log.md
|
|
25
|
+
```
|
|
26
|
+
|
|
27
|
+
Memory Bank files are not automatically versioned or shared across clones.
|
|
28
|
+
Check the repository's ignore/commit policy before relying on `_tasks/` records
|
|
29
|
+
or other files being tracked; `_tasks/` may be Git-ignored.
|
|
30
|
+
|
|
31
|
+
---
|
|
32
|
+
|
|
33
|
+
### PLAN
|
|
34
|
+
```markdown
|
|
35
|
+
# PLAN-<feature>
|
|
36
|
+
**Date:** YYYY-MM-DD **Status:** Awaiting Approval
|
|
37
|
+
|
|
38
|
+
## Goal
|
|
39
|
+
[One sentence]
|
|
40
|
+
|
|
41
|
+
## Phases
|
|
42
|
+
1. Phase 1 — @hermes
|
|
43
|
+
2. Phase 2 — @aphrodite
|
|
44
|
+
|
|
45
|
+
## Risks
|
|
46
|
+
- [Risk]
|
|
47
|
+
```
|
|
48
|
+
|
|
49
|
+
### IMPL
|
|
50
|
+
```markdown
|
|
51
|
+
# IMPL-<phase>-<agent>
|
|
52
|
+
**Date:** YYYY-MM-DD **Status:** Awaiting Themis Review
|
|
53
|
+
|
|
54
|
+
## What Was Implemented
|
|
55
|
+
- [file] — [what changed]
|
|
56
|
+
|
|
57
|
+
## Tests
|
|
58
|
+
- ✅ X tests / Coverage: Y%
|
|
59
|
+
```
|
|
60
|
+
|
|
61
|
+
### REVIEW
|
|
62
|
+
```markdown
|
|
63
|
+
# REVIEW-<feature>
|
|
64
|
+
**Status:** APPROVED | NEEDS_REVISION | FAILED
|
|
65
|
+
|
|
66
|
+
## Verdict
|
|
67
|
+
[APPROVED | NEEDS_REVISION | FAILED]
|
|
68
|
+
|
|
69
|
+
## Issues
|
|
70
|
+
- CRITICAL: X / HIGH: Y / MEDIUM: Z
|
|
71
|
+
```
|
|
@@ -84,6 +84,16 @@ placeholder.
|
|
|
84
84
|
|
|
85
85
|
---
|
|
86
86
|
|
|
87
|
+
## Routing
|
|
88
|
+
|
|
89
|
+
| Action / trigger | Reference |
|
|
90
|
+
|---|---|
|
|
91
|
+
| A delegation times out, you need a cooldown template or the tracking-table format, or you are about to checkpoint | [references/auto-continue-details.md](references/auto-continue-details.md) |
|
|
92
|
+
|
|
93
|
+
A missing reference interrupts the action — read nothing in its place, never improvise from memory.
|
|
94
|
+
|
|
95
|
+
---
|
|
96
|
+
|
|
87
97
|
## Details under demand
|
|
88
98
|
|
|
89
99
|
The cooldown templates, timeout/retry enforcement (the single authority is
|
|
@@ -35,40 +35,7 @@ biome check --write --unsafe <files>
|
|
|
35
35
|
|
|
36
36
|
## Review Checklist
|
|
37
37
|
|
|
38
|
-
|
|
39
|
-
- [ ] Logic is correct and complete
|
|
40
|
-
- [ ] Edge cases handled
|
|
41
|
-
- [ ] Error handling appropriate
|
|
42
|
-
- [ ] Performance acceptable
|
|
43
|
-
|
|
44
|
-
### Code Quality
|
|
45
|
-
- [ ] No duplication (DRY principle)
|
|
46
|
-
- [ ] Single responsibility functions
|
|
47
|
-
- [ ] Clear and descriptive naming
|
|
48
|
-
- [ ] Reasonable complexity (no god functions)
|
|
49
|
-
- [ ] Proper file sizes (<300 lines)
|
|
50
|
-
|
|
51
|
-
### Testing
|
|
52
|
-
- [ ] Tests cover the changed behavior where meaningful; documentation/format-only changes do not need invented tests
|
|
53
|
-
- [ ] Apply an explicit repository coverage threshold only when measurable and relevant to the changed code
|
|
54
|
-
- [ ] Integration tests where the change crosses workflow boundaries
|
|
55
|
-
- [ ] Relevant edge cases tested for the changed behavior/risk
|
|
56
|
-
- [ ] Relevant error conditions tested
|
|
57
|
-
|
|
58
|
-
### Security (OWASP Top 10)
|
|
59
|
-
- [ ] Input validation present
|
|
60
|
-
- [ ] No hardcoded secrets/credentials
|
|
61
|
-
- [ ] No XXE, CSRF, XSS vulnerabilities
|
|
62
|
-
- [ ] Authentication/authorization proper
|
|
63
|
-
- [ ] Sensitive data encrypted
|
|
64
|
-
- [ ] Rate limiting on sensitive endpoints
|
|
65
|
-
- [ ] Audit logging for security events
|
|
66
|
-
|
|
67
|
-
### Documentation
|
|
68
|
-
- [ ] Public functions documented
|
|
69
|
-
- [ ] Comments explain WHY not WHAT
|
|
70
|
-
- [ ] README/guides accurate
|
|
71
|
-
- [ ] API documentation complete
|
|
38
|
+
Run the itemized pass (Correctness, Code Quality, Testing, Security/OWASP, Documentation) and the SOLID check from [references/review-checklists.md](references/review-checklists.md) before issuing a verdict.
|
|
72
39
|
|
|
73
40
|
---
|
|
74
41
|
|
|
@@ -117,16 +84,6 @@ For comprehensive reviews, run 5 checks simultaneously:
|
|
|
117
84
|
|
|
118
85
|
---
|
|
119
86
|
|
|
120
|
-
## SOLID Principles Check
|
|
121
|
-
|
|
122
|
-
- **S**ingle Responsibility: Each class/function does one thing
|
|
123
|
-
- **O**pen/Closed: Extensible without modification
|
|
124
|
-
- **L**iskov Substitution: Subtypes behave like base types
|
|
125
|
-
- **I**nterface Segregation: Small, focused interfaces
|
|
126
|
-
- **D**ependency Inversion: Depend on abstractions, not concretions
|
|
127
|
-
|
|
128
|
-
---
|
|
129
|
-
|
|
130
87
|
## Anti-Patterns
|
|
131
88
|
|
|
132
89
|
- ❌ Reviewing entire file instead of diff
|
|
@@ -134,3 +91,13 @@ For comprehensive reviews, run 5 checks simultaneously:
|
|
|
134
91
|
- ❌ Ignoring security implications
|
|
135
92
|
- ❌ Vague feedback ("looks good", "needs work")
|
|
136
93
|
- ❌ Nitpicking style over substance
|
|
94
|
+
|
|
95
|
+
---
|
|
96
|
+
|
|
97
|
+
## Routing
|
|
98
|
+
|
|
99
|
+
| Action / trigger | Reference |
|
|
100
|
+
|---|---|
|
|
101
|
+
| Executing the itemized review pass (Correctness, Quality, Testing, Security, Documentation) or the SOLID check | [references/review-checklists.md](references/review-checklists.md) |
|
|
102
|
+
|
|
103
|
+
A missing reference interrupts the action — read nothing in its place, never improvise from memory.
|
|
@@ -0,0 +1,54 @@
|
|
|
1
|
+
# Code Review — Itemized Checklists and SOLID
|
|
2
|
+
|
|
3
|
+
Detail companion for `SKILL.md`. Read on demand when executing the itemized
|
|
4
|
+
review pass. The review scope, automated-check-first rule, feedback format,
|
|
5
|
+
severity definitions, and anti-patterns stay in `SKILL.md`.
|
|
6
|
+
|
|
7
|
+
---
|
|
8
|
+
|
|
9
|
+
## Review Checklist
|
|
10
|
+
|
|
11
|
+
### Correctness
|
|
12
|
+
- [ ] Logic is correct and complete
|
|
13
|
+
- [ ] Edge cases handled
|
|
14
|
+
- [ ] Error handling appropriate
|
|
15
|
+
- [ ] Performance acceptable
|
|
16
|
+
|
|
17
|
+
### Code Quality
|
|
18
|
+
- [ ] No duplication (DRY principle)
|
|
19
|
+
- [ ] Single responsibility functions
|
|
20
|
+
- [ ] Clear and descriptive naming
|
|
21
|
+
- [ ] Reasonable complexity (no god functions)
|
|
22
|
+
- [ ] Proper file sizes (<300 lines)
|
|
23
|
+
|
|
24
|
+
### Testing
|
|
25
|
+
- [ ] Tests cover the changed behavior where meaningful; documentation/format-only changes do not need invented tests
|
|
26
|
+
- [ ] Apply an explicit repository coverage threshold only when measurable and relevant to the changed code
|
|
27
|
+
- [ ] Integration tests where the change crosses workflow boundaries
|
|
28
|
+
- [ ] Relevant edge cases tested for the changed behavior/risk
|
|
29
|
+
- [ ] Relevant error conditions tested
|
|
30
|
+
|
|
31
|
+
### Security (OWASP Top 10)
|
|
32
|
+
- [ ] Input validation present
|
|
33
|
+
- [ ] No hardcoded secrets/credentials
|
|
34
|
+
- [ ] No XXE, CSRF, XSS vulnerabilities
|
|
35
|
+
- [ ] Authentication/authorization proper
|
|
36
|
+
- [ ] Sensitive data encrypted
|
|
37
|
+
- [ ] Rate limiting on sensitive endpoints
|
|
38
|
+
- [ ] Audit logging for security events
|
|
39
|
+
|
|
40
|
+
### Documentation
|
|
41
|
+
- [ ] Public functions documented
|
|
42
|
+
- [ ] Comments explain WHY not WHAT
|
|
43
|
+
- [ ] README/guides accurate
|
|
44
|
+
- [ ] API documentation complete
|
|
45
|
+
|
|
46
|
+
---
|
|
47
|
+
|
|
48
|
+
## SOLID Principles Check
|
|
49
|
+
|
|
50
|
+
- **S**ingle Responsibility: Each class/function does one thing
|
|
51
|
+
- **O**pen/Closed: Extensible without modification
|
|
52
|
+
- **L**iskov Substitution: Subtypes behave like base types
|
|
53
|
+
- **I**nterface Segregation: Small, focused interfaces
|
|
54
|
+
- **D**ependency Inversion: Depend on abstractions, not concretions
|
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: instruction-forge
|
|
3
|
+
description: "Write and maintain AI instructions — AGENTS.md, agent prompts, and skill packages. Use when creating or editing any directive artifact, deciding where a rule lives, or testing that a rule fires."
|
|
4
|
+
---
|
|
5
|
+
|
|
6
|
+
# Instruction Forge
|
|
7
|
+
|
|
8
|
+
> **The smallest context that fully specifies the required behavior. Every rule has exactly one home, and every rule is tested for where it fires — and where it must not.**
|
|
9
|
+
|
|
10
|
+
---
|
|
11
|
+
|
|
12
|
+
## When this fires
|
|
13
|
+
|
|
14
|
+
- Writing or editing any directive artifact: AGENTS.md (either artifact), an agent prompt, an instruction file, a SKILL.md.
|
|
15
|
+
- A rule keeps being repeated in prompts or conversation — it needs a single home.
|
|
16
|
+
- A directive changed and you must prove nothing the old text required was lost.
|
|
17
|
+
- Deciding WHERE a new rule lives at all — or concluding it should not be written down.
|
|
18
|
+
|
|
19
|
+
Do NOT fire for one-shot task instructions given in conversation; those stay in conversation. Only durable, reusable behavior gets an artifact.
|
|
20
|
+
|
|
21
|
+
## Where a rule lives — one home per fact
|
|
22
|
+
|
|
23
|
+
| Rule kind | Home | Loaded when |
|
|
24
|
+
|---|---|---|
|
|
25
|
+
| Repo setup (commands, layout, gates of THIS repo) | `AGENTS.md` managed region | Every session in the repo |
|
|
26
|
+
| Repo-agnostic behavior every session must pay for | `AGENTS.global.md` / shared instruction (`agents: ["*"]`) | Every session |
|
|
27
|
+
| Behavior scoped to one agent | `src/instructions/*.instructions.md` with `agents:` selector | That agent's prompt, at build |
|
|
28
|
+
| A recurring workflow a specialist executes | A skill (`SKILL.md`, ≤120-line core + `references/`) | On demand via grant |
|
|
29
|
+
| A runtime gate the plugin owns | `behaviors:` registry in `routing.yml` (then env/override), never prose | Plugin setup |
|
|
30
|
+
| Generated region content | `scripts/build-agents-md.mjs` (source of truth) — NEVER hand-edited | Regenerated |
|
|
31
|
+
|
|
32
|
+
Two artifacts, zero overlap: the project `AGENTS.md` carries repo setup; `AGENTS.global.md` carries repo-agnostic rules. A session pays for global + project exactly once — a rule in both is a defect (`tests/eval/instruction-golden.test.mjs` pins this).
|
|
33
|
+
|
|
34
|
+
## The contract (before any edit)
|
|
35
|
+
|
|
36
|
+
1. **Home first** — place the rule with the table above; if no home fits, the rule is not durable enough to write down.
|
|
37
|
+
2. **Nothing lost** — diff the old text's REQUIREMENTS against the new: every imperative, threshold, and exception either survives or has a named successor.
|
|
38
|
+
3. **Test the edges** — write the trigger case, the no-trigger case, and the exception. Structural cases go in `tests/eval/`; semantic cases are documented in the artifact.
|
|
39
|
+
4. **Budget** — the main artifact pays context every session; depth moves to `references/` behind a Routing table. Passing every check still does not GUARANTEE an agent obeys — say so, and keep enforcement deterministic (hooks, gates) where it matters.
|
|
40
|
+
|
|
41
|
+
## Routing
|
|
42
|
+
|
|
43
|
+
| Action / trigger | Reference |
|
|
44
|
+
|---|---|
|
|
45
|
+
| Writing or editing AGENTS.md or a shared instruction | [references/agents-md-doctrine.md](references/agents-md-doctrine.md) |
|
|
46
|
+
| Writing or editing a SKILL.md or skill package | [references/skill-doctrine.md](references/skill-doctrine.md) |
|
|
47
|
+
| Deciding how a rule is tested (trigger / no-trigger / exception) | [references/eval-doctrine.md](references/eval-doctrine.md) |
|
|
48
|
+
|
|
49
|
+
A missing reference interrupts the action — read nothing in its place, never improvise from memory.
|
|
50
|
+
|
|
51
|
+
## Rules
|
|
52
|
+
|
|
53
|
+
- Generated regions (`pantheon:generated`) are machine-written: change the generator, never the output.
|
|
54
|
+
- Existing conventions of the codebase outrank personal preference — a rule that fights the grain gets rejected, not encoded.
|
|
55
|
+
- Frontmatter contracts are enforced by tests (`tests/agent-skills-contract.test.mjs`); never add unlisted keys.
|
|
56
|
+
- An instruction that says "always" but has a known exception must state the exception or it is a defect, not a style choice.
|
|
57
|
+
- Route friction capture through `reflect` (it decides an artifact is warranted); this skill executes the writing.
|
|
@@ -0,0 +1,50 @@
|
|
|
1
|
+
# Agents.md Doctrine
|
|
2
|
+
|
|
3
|
+
The two artifacts are generated, disjunct, and budgeted. Every session pays
|
|
4
|
+
for global + project exactly once; overlap is a defect pinned by
|
|
5
|
+
`tests/eval/instruction-golden.test.mjs`.
|
|
6
|
+
|
|
7
|
+
## The managed region
|
|
8
|
+
|
|
9
|
+
- Sentinels: `<!-- pantheon:generated:begin -->` … `<!-- pantheon:generated:end -->`.
|
|
10
|
+
- Content between them is machine-written by `scripts/build-agents-md.mjs`
|
|
11
|
+
(source of truth: project sections in that script + `src/instructions/`).
|
|
12
|
+
- To change generated content: edit the generator or the instruction files,
|
|
13
|
+
then run `node scripts/build-agents-md.mjs`. The `--check` flag fails on a
|
|
14
|
+
stale artifact — CI and doctor both run it.
|
|
15
|
+
- User content outside the region is preserved verbatim across regeneration
|
|
16
|
+
(`writeAgentsMdManaged` carves by sentinel). The installer deliberately
|
|
17
|
+
excludes AGENTS.md from the file inventory: a file carrying user content is
|
|
18
|
+
never safe to claim or remove.
|
|
19
|
+
|
|
20
|
+
## Global vs project
|
|
21
|
+
|
|
22
|
+
- **Global artifact** (`AGENTS.global.md`, installed to the config dir as the
|
|
23
|
+
user's AGENTS.md): agent table + shared instruction bodies
|
|
24
|
+
(`agents: ["*"]`). Repo-agnostic — the same on every project.
|
|
25
|
+
- **Project artifact** (`AGENTS.md`): THIS repo's setup only — commands,
|
|
26
|
+
layout, lint/test gates, conventions. It must not contain a rule a session
|
|
27
|
+
could need on any other repo.
|
|
28
|
+
- A shared instruction is a file in `src/instructions/*.instructions.md` with
|
|
29
|
+
`agents: ["*"]`. An agent-scoped one declares `agents: [<id>]` and rides
|
|
30
|
+
that agent's prompt at build (`generateAgentPrompt`); it never rides
|
|
31
|
+
another agent's.
|
|
32
|
+
|
|
33
|
+
## Writing rules that survive
|
|
34
|
+
|
|
35
|
+
- One imperative per line where possible; a threshold is a number, not "fast".
|
|
36
|
+
- State the trigger ("when X"), the action ("do Y"), and the exception
|
|
37
|
+
("unless Z") — an "always" with a known unstated exception is a defect.
|
|
38
|
+
- No narrative, no rationale paragraphs: the artifact pays per session, the
|
|
39
|
+
rationale lives in docs/ or the commit that introduced the rule.
|
|
40
|
+
- Budget awareness: `logicalLineCount` and the carve tests bound the
|
|
41
|
+
artifacts; if growth is needed, carve a new shared instruction instead of
|
|
42
|
+
inflating a static section.
|
|
43
|
+
- Never restate what a hook or gate already enforces deterministically —
|
|
44
|
+
point at it once. Enforcement lives in code; prose only explains.
|
|
45
|
+
|
|
46
|
+
## Testing
|
|
47
|
+
|
|
48
|
+
Structural golden cases run in `tests/eval/instruction-golden.test.mjs`
|
|
49
|
+
(carve, zero-overlap, scoped routing). When you change a shared instruction
|
|
50
|
+
or the generator, add the negative case you just protected against there.
|
|
@@ -0,0 +1,37 @@
|
|
|
1
|
+
# Eval Doctrine — how a rule is tested
|
|
2
|
+
|
|
3
|
+
Instructions are code written in prose: they get trigger tests, no-trigger
|
|
4
|
+
tests, and exception tests. What they do NOT get is a guarantee — passing
|
|
5
|
+
every check proves the artifact says the right thing, never that an agent
|
|
6
|
+
will obey. Say so when reporting; keep real enforcement deterministic.
|
|
7
|
+
|
|
8
|
+
## The three cases (write all three)
|
|
9
|
+
|
|
10
|
+
1. **Trigger** — an input where the rule MUST change the behavior. Assert the
|
|
11
|
+
visible effect (block emitted, section present, route chosen).
|
|
12
|
+
2. **No-trigger** — a near-miss where the rule MUST stay silent. This is the
|
|
13
|
+
case authors forget; most instruction bugs are over-firing.
|
|
14
|
+
3. **Exception** — the stated "unless" of the rule, exercised end to end.
|
|
15
|
+
An exception you cannot exercise is a rule you cannot enforce.
|
|
16
|
+
|
|
17
|
+
## Where the tests live
|
|
18
|
+
|
|
19
|
+
- **Structural (CI, deterministic)** — `tests/eval/*.test.mjs` over the real
|
|
20
|
+
generators and artifacts: carve semantics, zero-overlap, scoped routing,
|
|
21
|
+
budgets, lock integrity, reference graphs. Import the generator, call it,
|
|
22
|
+
assert the output; a golden case is a diff you refuse, not a snapshot you
|
|
23
|
+
bless.
|
|
24
|
+
- **Semantic (documented, not automated)** — for rules whose trigger is
|
|
25
|
+
judgment ("proportional effort"), write the three cases as examples IN the
|
|
26
|
+
artifact (wrong/right pairs beat paragraphs) and note the limits.
|
|
27
|
+
- **Model-in-the-loop (gated)** — routing/complexity scoring against held-out
|
|
28
|
+
cases follows `docs/reference/offline-evaluation.md`: frozen rubric, slot
|
|
29
|
+
labels, pre-registered acceptance. Never run it ad hoc.
|
|
30
|
+
|
|
31
|
+
## The honest disclaimer
|
|
32
|
+
|
|
33
|
+
A green eval suite means: the artifacts say what they must, the seams hold,
|
|
34
|
+
and the deterministic gates fire where designed. It does not measure
|
|
35
|
+
compliance. When a rule matters enough that non-compliance is expensive,
|
|
36
|
+
move it out of prose — hook, gate, permission, or the behaviors registry —
|
|
37
|
+
and test THAT.
|
|
@@ -0,0 +1,62 @@
|
|
|
1
|
+
# Skill Doctrine
|
|
2
|
+
|
|
3
|
+
A skill is a workflow a specialist executes on demand. The core pays its
|
|
4
|
+
context cost every invocation; depth must be earned by the trigger, not
|
|
5
|
+
carried by default.
|
|
6
|
+
|
|
7
|
+
## Anatomy (the enforced contract)
|
|
8
|
+
|
|
9
|
+
- `src/skills/<id>/SKILL.md` — the ONLY required file. Frontmatter allows
|
|
10
|
+
exactly `name`, `description`, `license`, `compatibility`, `metadata`
|
|
11
|
+
(`tests/agent-skills-contract.test.mjs` rejects anything else); `name` must
|
|
12
|
+
equal the directory id; `description` must be non-empty and work as a
|
|
13
|
+
router.
|
|
14
|
+
- Optional `references/*.md` — depth, loaded on demand by the agent's own
|
|
15
|
+
`read`. Every shipped reference MUST be cited in the core's `## Routing`
|
|
16
|
+
table; an uncited reference is dead weight (validate-routing warns).
|
|
17
|
+
- Optional `scripts/` — deterministic helpers, never prose in disguise.
|
|
18
|
+
|
|
19
|
+
## The description is the router
|
|
20
|
+
|
|
21
|
+
Selection reads the description, not the body. It must name the trigger
|
|
22
|
+
surfaced to the model ("Use when…") and the domain keywords for routing.
|
|
23
|
+
Write it as the ONE line a busy router sees among 22.
|
|
24
|
+
|
|
25
|
+
## Core budget: 120 lines
|
|
26
|
+
|
|
27
|
+
`tests/agent-skills-contract.test.mjs` pins the budget. The core keeps:
|
|
28
|
+
|
|
29
|
+
- frontmatter, one-line thesis, When-this-fires (triggers AND non-triggers),
|
|
30
|
+
- the operational contract (the rules that change behavior — format contracts
|
|
31
|
+
stay in core: the agent cannot pause to load a file mid-output),
|
|
32
|
+
- the `## Routing` table + the interruption rule.
|
|
33
|
+
|
|
34
|
+
Everything else — worked examples, long checklists, edge cases, fixtures —
|
|
35
|
+
moves to `references/`. The split precedent is `auto-continue`: 2,656 →
|
|
36
|
+
1,003 tokens core, details under demand.
|
|
37
|
+
|
|
38
|
+
## The Routing section (mandatory when references/ exists)
|
|
39
|
+
|
|
40
|
+
```markdown
|
|
41
|
+
## Routing
|
|
42
|
+
|
|
43
|
+
| Action / trigger | Reference |
|
|
44
|
+
|---|---|
|
|
45
|
+
| <what the agent is about to do> | [references/x.md](references/x.md) |
|
|
46
|
+
|
|
47
|
+
A missing reference interrupts the action — read nothing in its place, never
|
|
48
|
+
improvise from memory.
|
|
49
|
+
```
|
|
50
|
+
|
|
51
|
+
The table is the source of truth; validate-routing fails the build when a
|
|
52
|
+
cited path is missing on disk and warns when a shipped reference is never
|
|
53
|
+
cited.
|
|
54
|
+
|
|
55
|
+
## Lifecycle
|
|
56
|
+
|
|
57
|
+
- New skill → add the catalog entry (`src/pantheon/pantheon-decision-catalog.ts`),
|
|
58
|
+
the index (`src/skills/README.md`), then run `node scripts/update-skills-lock.mjs`.
|
|
59
|
+
- Grants: an agent runs a skill only if its frontmatter `skills:` grants it;
|
|
60
|
+
permissions derive fail-closed (`permission.skill: { '*': 'deny' }`).
|
|
61
|
+
- Capture the friction through `reflect` first; if the artifact is an agent
|
|
62
|
+
(not a skill), route to `agent-forge` instead.
|
|
@@ -5,60 +5,13 @@ description: "Memory bank rules, structure, and optimization — complete guide
|
|
|
5
5
|
|
|
6
6
|
# Memory Bank
|
|
7
7
|
|
|
8
|
-
|
|
8
|
+
Pantheon memory bank — rules, structure, and maintenance. The rules below change behavior; tables of detail live in `references/`.
|
|
9
9
|
|
|
10
10
|
---
|
|
11
11
|
|
|
12
|
-
##
|
|
12
|
+
## Storage Guardrail
|
|
13
13
|
|
|
14
|
-
|
|
15
|
-
|--------|-------|-------|----------|---------|
|
|
16
|
-
| **Durable Memory** | `pantheon-memory` MCP (SQLite `memory.db`) | Agents with the relevant tool grant | No entry TTL; retained until explicitly forgotten | Reusable facts and decisions |
|
|
17
|
-
| **Persistence** | `pantheon-persistence` MCP | Agents with the relevant tool grant | Optional, separate checkpoint/KV store; checkpoint TTLs apply | Explicit session/workflow checkpoints and temporary state |
|
|
18
|
-
| **Memory Bank** | `.pantheon/memory-bank/` | Team | Files persist in the working tree; Git tracking depends on repository ignore/commit policy | Project context: architecture, patterns, progress, ADRs |
|
|
19
|
-
|
|
20
|
-
Pantheon does not implement `/memories/repo/` or `/memories/session/` stores.
|
|
21
|
-
Memory Bank files are read explicitly or on demand through
|
|
22
|
-
`pantheon://memory-bank/{path}` from `pantheon-resources`; they are not
|
|
23
|
-
auto-loaded. Memory namespaces are organizational filters, not access-control
|
|
24
|
-
boundaries, and tools are available only when the active agent is granted them.
|
|
25
|
-
|
|
26
|
-
---
|
|
27
|
-
|
|
28
|
-
## Memory Bank Structure
|
|
29
|
-
|
|
30
|
-
```
|
|
31
|
-
.pantheon/memory-bank/
|
|
32
|
-
├── 00-project.md ← What is this project? (fill once)
|
|
33
|
-
├── 00-architecture.md ← System design and agent hierarchy
|
|
34
|
-
├── 00-components.md ← Component breakdown and ownership
|
|
35
|
-
├── 00-tech-stack.md ← Tech stack, setup, environment
|
|
36
|
-
├── 01-active-context.md ← Current sprint focus, decisions, blockers (MOST IMPORTANT)
|
|
37
|
-
├── 02-progress-log.md ← What works, what's left, milestones (append-only)
|
|
38
|
-
├── _tasks/
|
|
39
|
-
│ ├── _index.md ← Task master list
|
|
40
|
-
│ └── TASK0001-name.md ← Individual task records
|
|
41
|
-
└── _notes/
|
|
42
|
-
├── _index.md ← Notes index
|
|
43
|
-
└── NOTE0001-topic.md ← Architectural decisions, findings
|
|
44
|
-
```
|
|
45
|
-
|
|
46
|
-
Memory Bank files are not automatically synchronized across clones. In
|
|
47
|
-
particular, `_tasks/` records may be Git-ignored; only files tracked and
|
|
48
|
-
committed under the repository's policy are shared through Git.
|
|
49
|
-
|
|
50
|
-
### File Update Frequency
|
|
51
|
-
|
|
52
|
-
| File | Fill when | Update frequency |
|
|
53
|
-
|------|-----------|-----------------|
|
|
54
|
-
| `00-project.md` | Project start | Rarely |
|
|
55
|
-
| `00-architecture.md` | Project start | On significant changes |
|
|
56
|
-
| `00-components.md` | Project start | When components added/removed |
|
|
57
|
-
| `00-tech-stack.md` | Project start | On stack changes |
|
|
58
|
-
| `01-active-context.md` | Each sprint | Each sprint / major decision |
|
|
59
|
-
| `02-progress-log.md` | First completion | Append-only per milestone |
|
|
60
|
-
| `_tasks/` | Sprint tracking | Per task |
|
|
61
|
-
| `_notes/` | Significant findings | Per finding |
|
|
14
|
+
Pantheon has no `/memories/repo/` or `/memories/session/` stores. Memory Bank files under `.pantheon/memory-bank/` are read explicitly or on demand via `pantheon://memory-bank/{path}` — never auto-loaded — and are not automatically synchronized across clones; Git tracking follows the repository's ignore/commit policy.
|
|
62
15
|
|
|
63
16
|
---
|
|
64
17
|
|
|
@@ -107,45 +60,13 @@ targets, rewrite the archive automatically, or create one artifact per phase.
|
|
|
107
60
|
|
|
108
61
|
---
|
|
109
62
|
|
|
110
|
-
##
|
|
63
|
+
## Routing
|
|
111
64
|
|
|
112
|
-
|
|
113
|
-
|
|
114
|
-
|
|
115
|
-
|
|
116
|
-
|
|
65
|
+
| Action / trigger | Reference |
|
|
66
|
+
|---|---|
|
|
67
|
+
| Needing where each memory system stores data, ownership, or lifetime | [references/storage-model.md](references/storage-model.md) |
|
|
68
|
+
| Creating/updating files and needing the layout or fill/update cadence | [references/structure-and-frequency.md](references/structure-and-frequency.md) |
|
|
69
|
+
| Unsure whether a write pattern (session files, phase handoffs, duplication) is allowed | [references/anti-patterns.md](references/anti-patterns.md) |
|
|
70
|
+
| Setting up the memory bank in a product repository | [references/product-adoption.md](references/product-adoption.md) |
|
|
117
71
|
|
|
118
|
-
|
|
119
|
-
@mnemosyne Append to 02-progress-log.md: [summary]
|
|
120
|
-
```
|
|
121
|
-
|
|
122
|
-
### ❌ Mandatory handoff after every phase
|
|
123
|
-
```
|
|
124
|
-
# Wrong: After every phase → handoff to @mnemosyne
|
|
125
|
-
|
|
126
|
-
# Right: Agent appends to 01-active-context.md directly
|
|
127
|
-
At sprint close → explicit @mnemosyne invocation
|
|
128
|
-
```
|
|
129
|
-
|
|
130
|
-
### ❌ Duplicating information
|
|
131
|
-
```
|
|
132
|
-
# Wrong: duplicate the same fact in both durable Memory and a project document
|
|
133
|
-
|
|
134
|
-
# Right: Durable reusable fact → pantheon-memory (when warranted and granted)
|
|
135
|
-
Narrative context → 00-project.md (read on demand)
|
|
136
|
-
```
|
|
137
|
-
|
|
138
|
-
---
|
|
139
|
-
|
|
140
|
-
## Adopting in a Product
|
|
141
|
-
|
|
142
|
-
```bash
|
|
143
|
-
mkdir -p .pantheon/memory-bank/_tasks .pantheon/memory-bank/_notes
|
|
144
|
-
touch .pantheon/memory-bank/_tasks/.gitkeep .pantheon/memory-bank/_notes/.gitkeep
|
|
145
|
-
# @mnemosyne Initialize memory bank for this repository
|
|
146
|
-
```
|
|
147
|
-
|
|
148
|
-
Add to `.github/copilot-instructions.md`:
|
|
149
|
-
```markdown
|
|
150
|
-
Always read .pantheon/memory-bank/01-active-context.md before answering.
|
|
151
|
-
```
|
|
72
|
+
A missing reference interrupts the action — read nothing in its place, never improvise from memory.
|
|
@@ -0,0 +1,33 @@
|
|
|
1
|
+
# Memory Bank — Anti-Patterns
|
|
2
|
+
|
|
3
|
+
Detail companion for `SKILL.md`. Read on demand when unsure whether a write
|
|
4
|
+
pattern violates the memory-bank rules. The golden rules themselves stay in
|
|
5
|
+
`SKILL.md`.
|
|
6
|
+
|
|
7
|
+
---
|
|
8
|
+
|
|
9
|
+
### ❌ Session output as files
|
|
10
|
+
```
|
|
11
|
+
# Wrong
|
|
12
|
+
Create IMPLEMENTATION_SUMMARY.md
|
|
13
|
+
Create STATUS.md
|
|
14
|
+
|
|
15
|
+
# Right
|
|
16
|
+
@mnemosyne Append to 02-progress-log.md: [summary]
|
|
17
|
+
```
|
|
18
|
+
|
|
19
|
+
### ❌ Mandatory handoff after every phase
|
|
20
|
+
```
|
|
21
|
+
# Wrong: After every phase → handoff to @mnemosyne
|
|
22
|
+
|
|
23
|
+
# Right: Agent appends to 01-active-context.md directly
|
|
24
|
+
At sprint close → explicit @mnemosyne invocation
|
|
25
|
+
```
|
|
26
|
+
|
|
27
|
+
### ❌ Duplicating information
|
|
28
|
+
```
|
|
29
|
+
# Wrong: duplicate the same fact in both durable Memory and a project document
|
|
30
|
+
|
|
31
|
+
# Right: Durable reusable fact → pantheon-memory (when warranted and granted)
|
|
32
|
+
Narrative context → 00-project.md (read on demand)
|
|
33
|
+
```
|