PRISM v2.8.0

v2.8.0 is a MINOR over v2.7.0 — corpus-access dispatch Phase 2: the lens-anchored auto-trigger. It turns the recommended_sources: field shipped at v2.7.0 into a live, advisory trigger — orchestration recognizes when a recommended source answers the question being worked, and proposes a lookup.

Surface

• Recognition (automatic). A lens carrying a recommended_sources: entry is trigger-capable. Orchestration recognizes, on its own, the conjunction: the lens is in play, its material question is being actively worked, and a recommended source's answers: binding matches that question. Recognition is the only thing that becomes automatic.
• Dispatch (advisory; operator-ratified). Recognition produces a candidate dispatch, not a dispatch. Orchestration shapes the investigation-posture Envelope and surfaces it for ratification; the operator ratifies, edits, or skips — silence does nothing. SP-9 ("silence is never consent") is preserved by keeping the autonomy in the recognition, never the dispatch; the trigger reuses the existing propose-then-ratify (P0→P1 Layer-3 ratification) posture and introduces no new consent model.
• Auto-populate. The matched source record auto-populates the Envelope's mandatory Source framing and Temporal frame caveats — the two fields Phase 1 made mandatory-but-hand-written — so the caveat is structurally impossible to silently omit. This closes the bidirectional self-containment loop: the question goes out self-contained, and now the outbound caveat travels automatically.
• Candidate logging + noise guard. Surfaced-but-unratified candidates are What's next material, not Dispatch-register entries. Noise guard: surface once per {lens, source, material-question} per engagement; record a decline; re-surface only on a material change to the question (the saturation predicate three-layer readiness already uses for Library coverage).

Additive and backward-compatible: engagements not using the trigger are unaffected. No lens-schema or lint-catalog change — the consumed field shipped at v2.7.0, and PRISM-LINT-08 remains reserved.

Phase status

• Phase 1 (v2.6.0) — manual and vendor-executed paths, manual invocation.
• Phase 2 (this release) — lens-anchored auto-trigger.
• Phase 3 — cowork-mcp auto-driven authenticated-session lookup; reserved, gated on the Cowork substrate maturing.

Calibration / report-back

• Trigger precision — how often recognition surfaces a genuinely useful candidate vs. noise across real engagements; the material-question match heuristic is rev. 1.
• Noise-guard granularity — whether once-per-{lens, source, material-question} with material-change re-surface over- or under-surfaces in practice.
• Auto-populate fidelity — whether the source-record framing:/recency: text reads well as Envelope caveats verbatim, or wants per-dispatch tailoring.

Version-pinned snapshot at this tag: PRISM_v2_8_0.md (byte-identical to PRISM.md).

PRISM v2.7.0 — recommended-sources-on-lens

PRISM v2.7.0 — recommended-sources-on-lens (corpus-access Phase 2, Build 1)A MINOR over v2.6.0. The Lens Library gains an optional recommended_sources: field that attaches a framework-curated list of external reference sources — each with a mandatory framing and recency caveat — to a lens's material question.### What's new- recommended_sources: field on the lens schema. Optional and backward-compatible, exactly like rubric_anchor: — entries without it remain valid. Each source record carries source, kind (narrative | structured-record), access (open-web | operator-authenticated), a mandatory framing: caveat, a mandatory source-scoped recency: posture, and an answers: binding to the lens's material question.- Populated on two high-yield lenses: - LL-D-008 "Compared to what?" (competitive substitution) - LL-D-009 "Does it pay back?" (commercial viability) - Each carries three sources: ideas.rip, pitch-deck libraries, and the CB Insights / PitchBook / Tracxn structured-record group.- Embedded Lens Library (Appendix G) and standalone mirror both advance to v0.12.- PRISM-LINT-08 (lens-recommended-sources-valid) reserved; lint catalog version 1 → 2.### No behavior changeThis release ships the field only. It is additive — there is no runtime behavior change. Auto-population of the corpus-access Envelope from this field is the consuming Phase-2 build that follows.### Artifacts- PRISM.md — always-current singleton (framework + embedded Lens Library v0.12 + skill frontmatter)- PRISM_v2_7_0.md — version-pinned snapshot, byte-identical to PRISM.md at this tag- lens/PRISM_lens_library.md — standalone Lens Library v0.12 (tag prism-lens-v0.12)PRISM v1.10.4 remains terminal on the v1.x line.

PRISM v2.6.0

v2.6.0 is a MINOR over v2.5.2. It adds corpus-access dispatch (Phase 1) — a first-class investigation posture for retrieving from named sources and document corpora, kept structurally distinct from epistemic triangulation.

What's new

• Posture made structural. The Execution Envelope (§3.2.1) now carries a Posture: field. Epistemic-posture Envelopes carry a dispatch shape; investigation-posture Envelopes carry none of {dispatch shape, vendor list, dispatch rationale}. The SP-15 triangulation guardrail now lives in Envelope structure, not just in prose.
• Corpus-access Envelope (new §4.13). A real spec for "go look this up in a named source": Source, Corpus kind, Source access, Driver, Fan, Tools, Question, Extract, Return form, Archive, mandatory Source framing and Temporal frame, and Return handling. Path routing: open-web → vendor-executed by default; operator-authenticated → cowork-mcp (vendor-executed barred — a hard credential boundary).
• Coverage fan. Fan: coverage (N) returns a recall-merge plus a retrieval-consistency note — explicitly not §4.3 Vendor Triangulation. §4.3 is guarded to fire only on epistemic-posture equivalence dispatch.
• Exhibits manifest (§3.2.3). Bundle return = canonical signed Master + companion artifacts. The manifest is mandatory whenever Archive: is requested, with per-entry provenance: filename · source · capture date · query answered · framing + temporal caveat. SP-14 gains a companion-file convention [project]_[promptID]_[source]_[seq].{png,pdf,xlsx}.
• Self-containment is bidirectional for corpus-access (§4.12): the Question goes out self-contained; the Return comes back caveat-attached.

Phase status

Phase 1 makes the manual and vendor-executed lookup paths operational. The cowork-mcp path is defined but reserved (alongside auto_drive). Still to come: Phase 2 (lens-anchored auto-trigger, gated on recommended-sources-on-lens) and Phase 3 (cowork-mcp auto-driven lookup).

Watch in use / calibration

• Whether the Posture: field reads cleanly at dispatch time or wants a shorter default.
• Whether the three-value Driver: enum (capability-local, distinct from the Axis-2 execution-driver enum) holds up once real operator lookups run, or wants the auto_drive-style mapping made more explicit.
• Whether coverage-fan returns stay legibly separate from triangulation in the Master, or need a stronger visual marker.
• Exhibits-manifest friction: is per-entry provenance the right granularity, or too heavy for small archives?

Design provenance

Decisions D1–D12: PRISM-workshop/design/corpus_access_dispatch_dd_rev1.md (private workshop repo).

Lint: 0 error, 57 info. Lens Library unchanged (v0.11).

PRISM v2.5.2

PATCH over v2.5.1. A documentation clarification — no functional change.

The repo_backed persisted-state set is now stated as a principle: the repo work folder holds the engagement's whole state — everything the work touches, inputs in and outputs out — with the enumeration explicitly illustrative rather than closed.

Operator-supplied Inputs are now named alongside the working artifacts and execution Outputs: the subject brief, reference documents, spreadsheets, decks, and information given in a chat prompt — captured at Setup or any time after. Information pasted into chat is written to a file rather than left in scrollback, since that is exactly the cross-surface state a different surface could not otherwise pick up — the loss repo_backed exists to prevent.

This is the inputs-side mirror of the v2.5.1 execution-returns clarification. Snapshot PRISM_v2_5_2.md byte-identical to PRISM.md at this tag. Lint 0 errors / 56 info (baseline-identical).

PRISM v2.5.1

PATCH over v2.5.0. A documentation clarification — no functional change.

§3.5.3 (the repo_backed mechanics) now states explicitly how execution results persist. Execution Outputs — the reports a vendor returns from a dispatched Envelope, whether Markdown, a Word/PDF document, or pasted text — save to the repo work folder the same way every other artifact does: the operator attaches the returned Output to the orchestration session (its normal lifecycle step), and orchestration commits it under the Claude-as-committer model. This works today with no extra machinery.

A future git-enabled execution session that commits its own returns directly is noted as a possible later direction (lower priority, not built) — the persistence-side parallel to the reserved auto_drive execution driver.

The mechanics step that lists persisted artifacts already included Outputs; v2.5.1 just makes the return path legible. Snapshot PRISM_v2_5_1.md byte-identical to PRISM.md at this tag. Lint 0 errors / 56 info (baseline-identical).

PRISM v2.5.0

MINOR over v2.4.0. Lands the mechanics for the persistence axis's repo_backed value, whose value and contract were fixed in v2.4.0.

What's new

persistence: repo_backed now has a documented build (new §3.5.3):

• Surface-agnostic by design. It earns its place even in a plain single_chat or projects session — the repo is where the Master and What's next survive across chats that would otherwise lose them to scrollback. Cross-surface switching (e.g. desktop Cowork ⇄ mobile Project, each picking up from the repo) is the additional payoff, not the baseline justification.
• 6-step Setup flow — repo + scoped PAT → work folder → engagement SI file → Project configuration → all artifacts (including the repo-resident What's next) save to the repo → free surface-switching.
• Committer model — orchestration commits directly with the operator's PAT (Claude-as-committer); operator-as-committer is offered as the conservative fallback.
• Operator PAT hygiene — minimum scope, credential-store storage, inject-then-strip, rotation.
• Engagement-SI skeleton — the 8-section per-engagement operating document the flow generates.

What's next (§3.4) gains a repo-resident variant note: same artifact, repo-stored, serving as the cross-surface pickup point.

The default ephemeral cell and the triple execution contract (Envelope · Self-check · Output) are unchanged.

Calibration / report-back

• This is the first axis-value to get real mechanics; the engagement-SI skeleton and the work-folder layout (prism/<engagement-slug>/) are first-draft conventions worth pressure-testing on a real engagement.
• The Claude-as-committer default places an operator PAT in a Project credential surface — feedback on whether the hygiene guidance is sufficient in practice is welcome.

Provenance

Lint: 0 errors / 56 info (baseline-identical). Snapshot PRISM_v2_5_0.md is byte-identical to PRISM.md at this tag.

PRISM Lens Library v0.12 — recommended_sources

PRISM Lens Library v0.12 — recommended_sources field

A MINOR over v0.11. Adds the optional recommended_sources: field to the lens schema: a framework-curated list of external reference sources, each bound to the lens's material question and carrying a mandatory framing and recency caveat.

What's new

• recommended_sources: field. Same 23 lenses, same triggers. The field is additive and backward-compatible — entries without it remain valid (like rubric_anchor:), so v0.12 introduces no behavior change for existing lenses.
• Populated on two high-yield lenses: LL-D-008 "Compared to what?" and LL-D-009 "Does it pay back?" — each with three sources (ideas.rip, pitch-deck libraries, and the CB Insights / PitchBook / Tracxn structured-record group).

Sub-schema

Each source record carries source, kind (narrative | structured-record), access (open-web | operator-authenticated), a mandatory framing: bias/handling caveat, a mandatory source-scoped recency: posture, and an answers: binding to the lens's material question.

Artifacts

• lens/PRISM_lens_library.md — always-current standalone catalog
• lens/PRISM_lens_library_v0_12.md — version-pinned snapshot, byte-identical at this tag

This library is also embedded in PRISM.md as Appendix G; the embedded copy advances to v0.12 in the same release (framework v2.7.0).

PRISM v2.4.0

MINOR over v2.3.1. Three-axis orchestration schema.

The single reserved execution_mode Setup flag is retired and replaced with three orthogonal axes, each a closed validated enum (per-axis halt-on-unrecognized preserved):

• Orchestration surface — single_chat (default) · projects · cowork
• Execution driver — manual (default; only built value) · auto_drive (reserved; gated to cowork)
• Persistence — ephemeral (default) · repo_backed

The default cell (single_chat / manual / ephemeral) reproduces v2.0 behavior exactly — no running project changes behavior. The triple execution contract (Envelope · Self-check · Output) is unchanged; all three axes describe the driver/config layer above it.

Also in this release

• Reserved-token re-homing: agentic_orchestration → the auto_drive driver value; automated_cross_vendor and plugin_equipped retired as tokens (the first decomposes into auto_drive × cross-vendor equivalence dispatch; the second is subsumed by the dispatch-time Tools: slot).
• SP-15 (Triangulation integrity) extended with the auto-drive boundary: auto-driving N distinct vendor apps is cross-vendor triangulation; auto-drive that collapses to single-vendor sub-agent fan-out is parallel execution, not triangulation.
• The Cowork surface capability set is documented as an open set (auto-drive, app-under-test, isolated-context Claude-seat execution, subagent investigation).

Lens Library unchanged at v0.11. Pinned snapshot: PRISM_v2_4_0.md (byte-identical to PRISM.md at this tag).

Calibration / report-back: the schema ships here; Cowork mode, repo-backed mechanics, and auto-drive skills are downstream items each shipped on their own. If the SP-15 auto-drive boundary feels ambiguous in any real cross-vendor run, report it.

PRISM v2.3.1

PATCH over v2.3.0 — reader legibility. Markup-only, no behavior change.

The two rubric-anchored lenses now carry their plain-language names at the prose sites where they previously appeared as bare IDs, so a reader no longer has to cross-reference the Lens Library to know what an anchor points at:

• LL-D-002 "Can anyone use?" — WCAG 2.2 (accessibility)
• LL-D-005 "Can attackers get in?" — OWASP ASVS 5.0.0 (security)

The embedded Lens Library, coverage-map listings, and changelog already named every lens at their mention sites and were left untouched.

No calibration items — this changes presentation only.

Snapshot: PRISM_v2_3_1.md (byte-identical to PRISM.md at this tag).

PRISM v2.3.0

PRISM v2.3.0 is a MINOR release refining a single Lens Library entry and bumping the embedded Library to v0.11.

What changed

• LL-D-008 ("Compared to what?") lens-binding refinement. The lens now names the category-vs-audience substitution trap explicitly. Its failure_mode calls out the common drift of mistaking category-bounded rival enumeration for substitute analysis; its minimum_scope_binding now requires naming the audience and the job before enumerating comparators, with substitutes defined by the job rather than the product category. A hardware device, a different workflow, or a do-nothing default can each be a substitute if it serves the same audience-job pair.
• Embedded Lens Library → v0.11. Appendix G and frontmatter lens_library_embedded move to v0.11 so default singleton-PRISM.md operators receive the refinement without attaching a standalone Library. (Lens track: prism-lens-v0.11.)

Additive posture-facilitation — no behavior break, no schema change, same 23 lenses.

Why a framework MINOR rather than a standalone lens bump

The fix needs to reach operators running the default singleton attachment, whose Library source is the embedded Appendix G copy. Bumping only the standalone would leave them on the embedded v0.10.

Motivation

A real audit converged on a false-uniqueness finding after enumerating category-bounded rivals while an audience-defined substitute — a different form factor serving the same job for the same audience — went unnamed. The refinement targets that failure posture.

Calibration / report-back

This is the lens's first real-world-driven revision; it remains pre-release (v1.0 awaits further calibration). If you apply LL-D-008 under v0.11 and the audience-first framing either catches a substitute you would have missed or feels like friction, that is exactly the signal worth reporting (Discussions → Ideas).

Version-pinned snapshot: PRISM_v2_3_0.md (byte-identical to PRISM.md at this tag). v1.x → v2 surface-drift map: Appendix D.